CONFLICTING CUES: THE ROLE OF RACE, GENDER, AND POLICY INFORMATION IN PRIMARY ELECTIONS. Eric Douglas Loepp

Transcription

1 CONFLICTING CUES: THE ROLE OF RACE, GENDER, AND POLICY INFORMATION IN PRIMARY ELECTIONS by Eric Douglas Loepp Bachelor of Arts in Leadership Studies, University of Richmond, 2008 Master of Arts in Political Science, University of Pittsburgh, 2012 Submitted to the Graduate Faculty of the Dietrich School of Arts and Sciences in partial fulfillment of the requirements for the degree of Doctor of Philosophy University of Pittsburgh 2015

2 UNIVERSITY OF PITTSBURGH DIETRICH SCHOOL OF ARTS AND SCIENCES This dissertation was presented by Eric Douglas Loepp It was defended on July 21, 2015 and approved by Kristin Kanthak, Associate Professor, Political Science Jon Hurwitz, Professor, Political Science Jonathan Woon, Associate Professor, Political Science Victoria Shineman, Assistant Professor, Political Science Melanie Hughes, Associate Professor, Sociology ii

3 Copyright by Eric Loepp 2015 iii

4 CONFLICTING CUES: THE ROLE OF RACE, GENDER, AND POLICY INFORMATION IN PRIMARY ELECTIONS Eric Loepp, Ph.D. University of Pittsburgh, 2015 This dissertation is fundamentally concerned with how individuals use information cues during primary elections to evaluate and select among a field of candidates all belonging to the same political party. When partisan heuristic cues well-established as the most critical determinant of vote choice are effectively held constant, voters are expected to turn to other easily accessible information about candidates in order to sort them and identify the most suitable option. This project assesses one such type of information the demographic status of candidates. Primary voters are expected to (1) prefer the most ideologically proximate candidate as their preferred party nominee and (2) employ ideological stereotypes embedded in demographic cues to help subtype and sort their primary options. Primary candidates, on the other hand, should appeal to different subgroups of voters via ideological signals embedded in policy messages presented to voters. Moreover, since certain primary candidates are considered counterstereotypical not striking voters as typical demographic groups associated with particular parties there are also possible gains and losses for candidates based purely on their demographic status and not merely the ideological tone of their messages. The interaction of different types of information should generate different preferences for various types of voters in primary elections. The first two chapters discuss demographic trends across America s main political iv

5 parties, discuss the void in the literature related to intra-party decision-making, and present a theory related to how both candidate and voter characteristics condition evaluations in a primary context. A third chapter details two survey experiments a low-information setting and a highinformation setting that are fielded to test theoretical expectations. Chapters Four through Six present the results of the two studies. A concluding chapter summaries the findings and integrates this work into the larger literature on cue use in an electoral context. I also discuss limits of the current project and specify a series of steps to more thoroughly probe the issues initially tested in this dissertation. v

6 TABLE OF CONTENTS 1.0 CHAPTER 1: INTRODUCTION WHY PRIMARIES THE (OLD) DILEMMA OF THE COUNTERSTEREOTYPICAL REPUBLICAN CANDIDATE PARTY DIFFERENCES LEAD TO AN EMERGING PUZZLE Can Democrats be Demographically Counterstereotypical? SUMMARY AND NEXT STEPS Project Outline CHAPTER TWO: THEORY USING (EASY) INFORMATION: RACE, GENDER, AND IDEOLOGICAL INFERENCE HOW VOTERS EVALUATE PRIMARY CANDIDATES: ASSUMPTIONS UNDERLYING INFORMATION USE SORTING AND SELECTING WITH EASY INFORMATION LEVELING THE PLAYING FIELD: INDIVIDUATING IDEOLOGICAL INFORMATION EXPECTANCY VIOLATIONS AND POSITIVE AFFECT Policy Information and Democratic Party Choice vi

7 2.6 SUMMARY AND LOOKING AHEAD CHAPTER THREE: METHODS AND DATA INVESTIGATION STRATEGY PARTICIPANT RECRUITMENT AND VALIDATION OF RESPONSES Subject Assignment EXPERIMENTAL TREATMENT Manipulating Demographic Cues Manipulating Policy Information KEY DEPENDENT VARIABLES OTHER VARIABLES SUMMARY CHAPTER FOUR: LOW-INFORMATION CONTEXT DEFINING AND CONTRASTING IDEOLOGUES AND PARTISANS CANDIDATE EVALUATIONS: REPUBLICANS Perceived Ideology: Republicans Ideological Congruence: Republicans Evaluating Candidates on Support and Representation Capacity: Republicans CANDIDATE EVALUATIONS: DEMOCRATS Perceived Ideology: Democrats Ideological Congruence: Democrats Evaluating Candidates on Support and Representation Capacity: Democrats vii

8 4.4 CONCLUSIONS: A CASE OF MINIMAL DEMOGRAPHIC EFFECTS CHAPTER 5: IDEOLOGICAL PERCEPTIONS IN A HIGH-INFORMATION CONTEXT MANIPULATION CHECK THE EFFECTS OF DEMOGRAPHIC CUES AND POLICY INFORMATION ON IDEOLOGICAL PERCEPTIONS: REPUBLICANS Assessing Race, Gender, and Policy Information on Perceived Candidate Ideology Policy Information, Gender, and Perceptions of Candidate Ideology Policy Information, Race, and Perceptions of Candidate Ideology Assessing Race, Gender, and Policy Information on Perceived Ideological Congruence Gender, Policy Information, and Ideological Congruence Assessing Race and Policy Information on Perceived Ideological Congruence THE EFFECTS OF DEMOGRAPHIC CUES AND POLICY INFORMATION ON IDEOLOGICAL PERCEPTIONS: DEMOCRATS CONCLUSIONS: POLICY INFORMATION SHAPES IDEOLOGICAL PERCEPTIONS CHAPTER SIX: ATTITUDES TOWARDS CANDIDATES viii

9 6.1 THE EFFECTS OF DEMOGRAPHIC CUES AND POLICY INFORMATION ON CANDIDATE EVALUATIONS: REPUBLICANS Gender Effects Support Representation Capacity Racial Effects Support Representation Capacity CONCLUSIONS: SOME EVIDENCE OF EXPECTANCY VIOLATIONS CHAPTER SEVEN: DISCUSSION AND CONCLUSION KEY FINDINGS Voters Candidates REVISING THE REPUBUBLCAN DILEMMA WHAT ARE THE ADVANTAGES TO BEING COUNTERSTEREOTYPICAL? IMPLICATIONS FOR REPUBLICAN PARTY MOVING FORWARD CONTRIBUTIONS TO THE LITERATURE, LIMITATIONS, AND NEXT STEPS Gendered and Racialized Policy Domains Viability, Strategic Voting, and Multi-Candidate Contests Reconciling Other Forms of Information Conflict ix

10 7.5.4 Clarifying Information-Based and Affective-Based Models of Candidate Evaluation APPENDIX A APPENDIX B APPENDIX C BIBLIOGRAPHY x

11 LIST OF TABLES Table 1: Number and Success Rate of Women in Congress by Party, Table 2: Summary Statistics of Studies Compared to Recent National Sample Table 3: Description of Study 1 Treatment Conditions Table 4: Description of Study 2 Treatment Conditions Table 5: Experimental Treatments Used in Study Table 6: Primary Dependent Variables Table 7: Ideological and Partisan Distribution of Study 1 Participants Table 8: Description of Variables in Regression Models Table 9: Determinants of Perceived Candidate Conservatism (Republicans) Table 10: Average Ideological Distance Between Voter and Candidate (Republicans) Table 11: Determinants of Perceived Ideological Congruity Between Voter and Candidate (Republicans) Table 12: Support for Candidate as a Function of Ideology (Republicans) Table 13: Perceptions of Representation Capacity as a Function of Ideology (Republicans) Table 14: Effects of Ideological Congruence on Perceptions of Support/Represent.Capacity (Republicans) Table 15: Determinants of Perceived Candidate Conservatism (Democrats) xi

12 Table 16: Average Ideological Distance Between Voter and Candidate (Democrats) Table 17: Determinants of Perceived Ideological Congruity Between Voter and Candidate (Democrats) Table 18: Support for Candidate as a Function of Ideology (Democrats) Table 19: Perceptions of Representation Capacity as a Function of Ideology (Democrats) Table 20: Effects of Ideological Congruence on Perceptions of Support/Represent.Capacity (Democrats) Table 21: Perceptions of Candidate Ideology Across Ideological Subgroups: Republicans Table 22: Perceptions of Candidate Ideology Across Ideological and Demographic Subgroups: Republicans Table 23: Perceptions of Candidate Ideology in Control Condition (Republicans) Table 24: Perceptions of Ideological Congruence in Control Condition (Republicans) Table 25: Perception of Male and Female Ideology by Policy Message Type (Republicans) Table 26: Perception of White and African American Ideology by Policy Message Type (Republicans) Table 27: Perceptions of Ideological Incongruence With Male and Female Candidates by Policy Message Type (Republicans) Table 28: Perception of Ideological Congruence with White and African American Candidates by Policy Message Type (Republicans) Table 29: Differences in Support for Candidates Across Ideological Subgroups (Republicans) 142 Table 30: Differences in Representation Capacity of Candidates Across Ideological Subgroups (Republicans) xii

13 Table 31: Effects of Gender, Ideology, and Ideological Incongruence on Support for Candidates Table 32: Effects of Gender, Ideology, and Ideological Incongruence on Perceived Representation Capacity Table 33: Effects of Race, Ideology, and Ideological Incongruence on Support for Candidates 156 Table 34: Effects of Race, Ideology, and Ideological Incongruence on Perceived Representation Capacity Table 35: Perceptions of Candidate Ideology in Control Condition (Democrats) Table 36: Perceptions of Ideological Congruence in Control Condition by Policy Message (Democrats) Table 37: Perception of Male and Female Candidate Ideology by Policy Message Type (Democrats) Table 38: Perception of White and African American Candidate Ideology by Policy Message Type (Democrats) Table 39: Perceptions of Ideological Congruence With Male and Female Candidates by Policy Message Type (Democrats) Table 40: Perception of Ideological Congruence with White and African American Candidates by Policy Message Type (Republicans) xiii

14 LIST OF FIGURES Figure 1: Targeted Voters In Primary and General Election Campaigns... 7 Figure 2: Female Success in Primaries Across Parties in U.S. House Elections, Figure 3: Party Identification by Demographic Group, Figure 4: Women's Representation in U.S. House by Party, Figure 5: Women's Representation in U.S. Senate by Party, Figure 6: Assignment of Subjects in Study Figure 7: Assignment of Subjects in Study Figure 8: Ideological and Partisan Distribution of Study 1 Participants Figure 9: Perceived Ideological Location of Republican Candidates Figure 10: Predicting Candidate Conservatism As a Function of Voter Ideology (Republicans) 79 Figure 11: Perceived Ideological Incongruence As a Function of Voter Ideology (Republicans) 83 Figure 12: Willingness to Support Candidate As a Function of Ideology (Republicans) Figure 13: Perceptions of Representation Capacity as a Function of Ideology (Republicans) Figure 14: Perceptions of Support/Rep. Capacity as a Function of Ideological Incongruence (Republicans) Figure 15: Perceptions of Support/Rep. Capacity as a Function of Ideological Congruence (Republicans) xiv

15 Figure 16: Perceived Ideological Location of Democratic Candidates (All Democrats) Figure 17: Predicting Candidate Conservatism As a Function of Voter Ideology (Democrats).. 97 Figure 18: Perceived Ideological Congruence As a Function of Voter Ideology (Democrats) Figure 19: Willingness to Support Candidate As a Function of Ideology (Democrats) Figure 20: Perceptions of Representation Capacity as a Function of Ideology (Democrats) Figure 21: Perceptions of Support/Rep. Capacity as a Function of Ideological Incongruence (Democrats) Figure 22: Perceptions of Support/Rep. Capacity as a Function of Ideological Incongruence (Democrats) Figure 23: Perceptions of Control Group Candidate Ideology by Policy Message (Republicans) Figure 24: Perceptions of Ideological Congruence in Control Condition by Policy Message (Republicans) Figure 25: Perception of Male and Female Ideology by Policy Message Type (Republicans) Figure 26: Perception of White and African American Ideology by Policy Message Type (Republicans) Figure 27: Perceptions of Ideological Incongruence by Gender and Policy Message Type (Republicans) Figure 28: Perception of Ideological Incongruence by Race and Policy Message Type (Republicans) Figure 29: Effects of Gender and Ideology on Support for Candidates Figure 30: Effects of Gender and Ideological Incongruence on Support for Candidates xv

16 Figure 31: Effects of Gender and Ideology on Perceived Representation Capacity of Candidates Figure 32: Effects of Gender and Ideological Incongruence on Perceived Represent. Capacity of Candidates Figure 33: Effects of Race and Ideology on Support for Candidates Figure 34: Effects of Race and Ideological Congruence on Support for Candidates Figure 35: Effects of Race and Ideology on Perceived Representation Capacity of Candidates 163 Figure 36: Effects of Race and Ideological Incongruence on Perceived Represent. Capacity of Candidates Figure 37: Perceptions of Control Group Candidate Ideology by Policy Message (Democrats) 218 Figure 38: Perceptions of Ideological Congruence in Control Condition (Democrats) Figure 39: Perception of Male and Female Ideology by Policy Message Type (Democrats) Figure 40: Perception of White and African American Ideology by Policy Message Type (Democrats) Figure 41: Perceptions of Ideological Congruence by Gender and Policy Message Type (Democrats) Figure 42: Perception of Ideological Incongruence by Race and Policy Message Type (Democrats) xvi

17 PREFACE I wish to extend my most heartfelt thanks and appreciation to the friends, family, and faculty who supported me during the completion of this project. xvii

18 1.0 CHAPTER 1: INTRODUCTION Following Barack Obama s re-election in the 2012 presidential election, the Republican Party began to more actively reflect on how to respond to an increasingly pressing shift in American demographics: white voters are comprising less and less of the overall voting population as other racial groups namely African Americans and Hispanics participate more in politics. Since non-white voters overwhelmingly support the Democratic Party, this evolution represents, at best, a legitimate electoral concern for Republican Party, and, at worst, an existential threat to the Party s long-term viability. Not only have Republicans struggled to appeal to new voters, but the party has even lost ground in some key areas for instance, in the 2004 presidential election George W. Bush collected forty-four percent of the Hispanic vote (nearly reaching parity with his Democratic opponent), but eight years later Mitt Romney received less than one-third of Hispanic votes. 1 The last few years have seen a rise in movements designed to address this demographic challenge. Internal party movements like the Future Majority Caucus were developed to 1 The Washington Post ( recently noted that if the 2016 Republican presidential nominee matches Mitt Romney s 2012 tally of 17 percent of the nonwhite vote, that candidate would have to attract support from 65 percent of white voters, a feat no achieved only by Ronald Reagan in his landslide 1984 re-election victory. George W. Bush managed to win 58 percent of the white vote and 26 percent of the nonwhite vote in 2004, yet given demographic shifts in the last decades, these same proportions would not be sufficient to achieve a Republican victory in

19 promote the election of Republican Hispanics and women to public office. 2 In December 2012, a month after Obama s re-election, Reince Priebus, Chairman of the Republican National Committee, announced a new initiative. The Growth and Opportunity Project (GAOP) was designed to provide an honest review of the 2012 election cycle and develop strategies to bolster the Republican Party s chances in future elections. 3 One of the most central goals of the GAOP is to diversify the party s constituency. As authors of the GAOP outline put it: Unless the (Republican National Committee) gets serious about tackling this problem, we will lose future elections. [ ] We have to work hard at engaging demographic partners and allies (Barbour et al. 2013, 12). Whit Ayres, Republican analyst and pollster for Senator Marco Rubio s 2016 election bid, echoes this sentiment in a recent book, submitting that if Republicans are to regain the White House, they must fare significantly better among minorities than they have in the past (Ayers 2015). The GAOP authors stress inclusion throughout their report, and of the fourteen specific recommendations they prescribe in the Demographic Partners section, five focus explicitly on targeting minority voters in an effort to expand the Party s base. Two others emphasize the need for the Party to recruit non-white and non-male candidates to run as Republicans. The authors note that Republican Party committees should be encouraging and championing their desire to seek elective office (Barbour et al. 2013, 21) Five co-chairs were assigned to the GAOP project Henry Barbour, Sally Bradshaw, Ari Fleischer, Zori Fonalledas, and Glenn McCall. Their methodology included the following: meeting or speaking with more than 2,600 people, both inside and outside Washington, and including voters, technical experts, private sector officials, Party members, [ ] elected office holders, [ ] Republicans from all ideological backgrounds; Holding in-depth focus groups with voters who used to call themselves Republicans but who left the Party because they thought we weren t conservative enough or because we were too conservative; conducting a poll among 2,000 Republican Hispanic voters; launching a survey of political practitioners at the state and national level as well as Republican and Independent pollsters; collected data from more than 36,000 individuals that participated in an online survey to determine priorities for the Republican Party. 2

20 On the whole, the GAOP outlines a number of strategies that, in theory, will help the Party achieve its goals. Yet largely absent from the pages of the GAOP report is a discussion of how Republican voters participating in primary elections will respond to efforts to reach out to non-white voters or how Republicans will react to female or minority candidates running under the Republican banner. 4 As will be discussed below, citizens are known to infer about political candidates on the basis of demographic features like race and gender. 5 Scholars have speculated that these tendencies will have a detrimental effect on counterstereotypical candidates those that do not conform to expectations or fit existing stereotypes like female or African American Republicans, most notably because voters tend to assume non-white and non-male candidates to be more ideologically liberal than white men (Koch 2000, McDermott 1997). I argue that primaries represent perhaps the greatest hurdle to the Republican Party s long-term demographic vision as it is currently conceived in the GAOP report, for it is when a non-white or non-male faces a conventional white male Republican in intra-party contests that Republican voters must be convinced to support candidates that do not look the part. How can counterstereotypical political candidates overcome this challenge? More generally, what are the consequences good or bad of being a non-white or non-male candidate in a primary election? Before proposing an answer to these questions, we must first understand the nature of the challenge itself. It starts with primary elections themselves. 4 GAOP s discussion of reforming primary elections focuses on logistical and scheduling changes for 2016 reducing the number of primary debates and moving up primary elections and ignore potential challenges involved in motivating Republican primary voters to support unconventional candidates. 5 It should be noted that I employ a variety of terms (such as sex and gender ) to indicate demographic differences between political candidates in order to help facilitate the flow of the narrative. However, as Virginia Prince (2005, 29) notes, Sex and gender are not the same thing. We are born [ ] not only into male and female, but into man and woman. In other words, female is an adjective, reflecting individuals sense of identity, while sex is a noun referencing anatomy. These two concepts overlap for many individuals, but not universally, and therefore this technical note is warranted. Nevertheless, this project is only concerned with how individuals use physical attributes (race, sex) to sort and evaluate political candidates, and therefore references in this text like gender, women, and female all refer to a candidate s sex. 3

21 1.1 WHY PRIMARIES While race and gender are known to be significant determinants in political impression formation generally, we should expect them to be particularly relevant in primary elections. Political research on demographic cues has historically ignored the role of parties (King and Matland 2003; Shafer 2013), yet even as parties are brought into the scholarly fold, investigations of the role of [demographic cues in] intra-party contests are rare (Jackman and Vavreck 2010, 155). Our collective focus on race and gender in general elections, while serving a crucial end in its own right, nevertheless skips over the electoral step in which race and gender ought to be even more central considerations in candidate evaluations primary elections where party labels are not the predominant factor in vote choice and should therefore be more substantively significant in our collective quest to uncover how voters form attitudes and make decisions. Several factors contribute to the centrality of demographic cues in these primary elections. For instance, from a purely technical standpoint these cues matter because candidates must win primary contests in order to compete in the fall general election and thus any effort to increase descriptive representation of women and racial minorities in office must start in the spring. Yet primary contests also matter from a behavioral perspective. While the Republican Party has concluded that it needs more female and minority candidates to run under the party banner, programs like the GAOP fail to account for potential challenges these candidates may encounter within the party simply by running to be the nominee in the fall. Authors of the GAOP, for example, acknowledge that the Party must recognize the unique challenges female candidates face in winning elections, but the recommendation associated with this observation implores the Party to provide training programs for potential female candidates that includes fundraising guidance, digital strategy, etc. (21). These prescriptions fail to account for how 4

22 Republican voters (potentially) evaluate unconventional candidates in other ways. The electoral landscape is also considerably more complicated during spring intra-party elections than during general elections in the fall. Primary elections represent extremely complex information environments. Unlike general elections, which feature one candidate from each major political party, the field of primary contenders often begins with many candidates. Since voters are known to minimize cognitive output in political decision making (Downs 1957; Lupia 1994), the extensive field of candidates in primary elections increases the likelihood that voters attempting to distinguish one candidate from another will reach for easily accessible information like race and gender rather than seek more appropriate but also more costly policy information. Indeed, voters are increasingly likely to rely on heuristic cues when facing a complex decision (Lau and Redlawsk 2001) and reliance on cues is known to decrease as information about candidates increases (Matson and Fine 2006). Recent work by Jones (2014) also finds that race affects impressions of candidates even when they provide other political information to voters. A related issue concerns the nature of choice in primary elections. The high degree of ideological similarity and policy congruence among candidates in primary elections dramatically reduces the cost of defecting from one candidate to another. As a result, voters who are particularly disposed to support or reject one candidate over another should have relatively little trouble doing so in a primary contest because alternative choices are ideologically proximate. In a general election, on the other hand, this defection is far more costly, for the alternative vote choice requires a voter to support another party (or abstain from the election altogether). To the extent that race and gender affect voters willingness to support political candidates, their substantive effects are much more relevant in spring primary elections. 5

23 A fourth reason to focus on race and gender explicitly in primary elections is due to the fact that the voting population shifts ideologically during spring nomination contests and the general election in the fall. Figure 1 provides an abstract illustration of this idea. In the general election, candidates focus on rallying their own partisan supporters, attracting independents, and picking off some moderate voters from the other party. In primary contests, on the other hand, the constituency to which candidates appeal is significantly smaller, focusing on only about half of the general ideological spectrum. 6 While the median voter in a general election represents the ideological center and may or may not affiliate with a political party, in primary elections the median voter falls in the middle of partisan distributions of Democrats and Republicans, respectively, and almost certainly belongs to one of the major parties. As a result, a voter s location relative to the rest of the voting public shifts, sometimes dramatically, depending on the election context. A Republican voter located just to the right of the ideological center, for instance, would be considered a moderate in the general election but would represent the liberal end of the spectrum among a population consistently only of Republicans. Given the tendency to impose an ideological penalty on female or African American candidates (e.g. to presume they are more liberal than white men), the ideological distribution of voters within party primaries implies that different types of partisan voters may be more or less likely to support some candidates over others. 6 It could be argued that in open-primary states where voters from all parties may select which party s primary they wish to participate in the voting population will not be comprised solely of citizens from a single partisan stripe. While there is some truth to this, there is evidence to suggest that candidates target the primary electorate (e.g. more ideologically extreme voters) rather than the overall median voter (Brady, Han, and Pope 2007). 6

24 Figure 1: Targeted Voters In Primary and General Election Campaigns The most crucial factor motivating this effort to study the role of race and gender in primary elections, however, is also the most basic: in primary elections the most important and oft-used electoral cue party affiliation is effectively neutralized. Scholars regularly affirm that a candidate s party identification provides the central cue that helps voters make an informed choice (Aldrich 1995; Mondak 1993; Schaffner and Streb 2002). Party labels dramatically simplify the voting process by allowing individuals to associate a general political/ideological philosophy with particular candidates without needing to invest heavily in learning about specific candidates (Rahn 1993). Yet in primary elections this cue is largely useless as a tool for distinguishing a voter s electoral options because all candidates affiliate with the same party. These unique features of primary elections imply that since voters share a partisan ingroup status with all candidates they must find some basis for discerning among them. As noted above, the lack of utility provided by partisan cues in primary elections suggests that voters will reach for other accessible cues like race and gender during primary contests. This renders demographic information far more relevant in a primary context than in a general election where party labels dominate the decision-making calculus. As a result, the potential for the Republican 7

25 Party not only to recruit but to nominate counterstereotypical candidates will depend in no small part on how its own voters process information about and evaluate Republican candidates in party primaries. But how do they do that, and why is it problematic for the very candidates currently championed by Republican leaders as essential for the longevity of the party? 1.2 THE (OLD) DILEMMA OF THE COUNTERSTEREOTYPICAL REPUBLICAN CANDIDATE Just as party labels are known to serve as heuristic devices that allow individuals to infer about political parties, demographic cues like race and gender affect how citizens evaluate politicians. Political scientists have long established voters latent propensity to impose a variety of traits and qualities on political candidates purely as a function of demographic status. Voters presume, for example, that non-white or non-male candidates possess certain traits that white men do not (Huddy and Terkildsen 1993), and to be particularly well- or poorly-suited to tackle various policy domains like education or national defense (Sigelman et al. 1995). As noted above, women (Koch 2002) and African Americans (Mcdermott 1997) are also perceived by individuals to be more ideologically liberal, all else equal, than white males. Koch (2000) for instance, uses pooled election data to compare where voters place male and female candidates on an ideological scale, controlling for a variety of other covariates, including estimates of actual candidate ideology. He concludes that utilization of gender stereotypes to infer candidates ideological orientation/position leads citizens to perceive female candidates as being more liberal than they are (426). McDermott (1998) utilizes an experiment in which fictional candidates without party labels are either male or female. She finds that more liberal (conservative) respondents are more 8

26 likely to express support for the female (male) candidate. Similarly, Republican respondents tend to prefer the male candidate while Democratic respondents favored the female candidate. 7 Most political office seekers possess a demographic status that is ideologically consistent with their partisan status. That is, the ideological leanings voters associate with each type of information are in the same direction. For instance, gender status female implies ideological liberalism, and so does the party label Democrat, and therefore a female Democrat poses little cognitive challenge to voters attempting to infer about her political views. White male Republicans are similarly stereotypical in this sense, for male and Republican are both associated with a conservative ideological output. Counterstereotypical candidates like female or African-American Republicans present a problem. They possess multiple cues demographic and partisan that do not align with preconceived notions concerning these candidates ideological disposition. Republican women, for instance, simultaneously provide a liberal cue (gender) along with a conservative cue (partisanship) (Dolan 2004; Koch 2000). The resulting confusion leads to a dilemma for counterstereotypical political candidates: voters may withdraw/fail to support them, not necessarily out of sexism or racism, but because they are not sure what conclusions to draw. When voters are unsure about a candidate, they are less likely to support her (Alvarez 1997). Furthermore, as discussed above, the primary context is one in which it is largely costless to switch one vote choice for another since candidates are so similar to one another to begin with, and therefore this sort of confusion is particularly dangerous for counterstereotypical candidates in this stage of the game. Recent election statistics are consistent with this classic formulation of this dilemma. 7 Interestingly, McDermott also includes a treatment condition in featuring only male candidates. In this scenario, respondents ideology has no effect on expressed vote choice. 9

27 Since there is not enough data to analyze racial minority groups running as Republicans, I focus for the moment on women, although the logic of the dilemma applies similarly to other counterstereotypical groups. Scholars note that female candidates, once nominated, can raise money and garner votes at similar rates as their male counterparts (Burrell 1998; Dolan 1998; Smith and Fox 2001). Seltzer, Newman, and Leighton (1997, 79) go as far as to say that winning elections has nothing to do with the sex of the candidate. Some work even finds that women enjoy similar primary election rates as men (Lawless and Pearson 2008). Yet when we focus on primary elections across parties, we see significant differences between stereotypical Democratic female candidates in their party primaries and counterstereotypical Republican female candidates in theirs. Consider, for instance, the midterm elections in 1994 and 2010, two years that saw large Republican gains in Congress. In both elections, there were more non-incumbent female Republican candidates running in party primaries than non-incumbent female Democratic candidates. In 1994, 37% of Republican women won the primaries they entered, compared to 48.7% of Democratic women. In 2010, the gap increased; 31.4% of Republican women won their primary elections compared to 50.3% of female Democrats. Two of the Republican Party s most successful elections in recent history, in short, were not good for Republican women, and it is at least in part because they did not make it onto the general election ballot. Looking at recent trends more generally, we see more evidence that counterstereotypical candidates struggle in primaries. Table 1 and Figure 2 present the win rates for Democratic women and Republican women in their respective primaries. For over twenty years Democratic women have succeeded in securing their party s nomination at higher rates than Republican women, and the difference between the parties in terms of female representation on the general 10

28 election ballot has increased since 2008 indeed, Democratic women have a statistically significant advantage over Republican women in terms of primary election win rate over the last decade. 8 It must be noted that these data do not distinguish incumbents from challengers. Given that women are more likely to run (and hold office) as Democrats, coupled with the fact that incumbents enjoy considerable advantages that aid in seat retention, the data may paint a somewhat more favorable portrait of electoral life as a female Democrat than a female Republican. Still, overall the descriptive statistics combined with the anecdotal data above suggest that Republican women struggle to secure their party s nomination moreso than do their Democratic counterparts. This poses a challenge to the Republican Party s vision of a more diverse candidate pool in future elections. 8 Democratic women enjoy a mean win rate during this period of 53.92% to Republican women s mean win rate of 44.14% (t=5.12, p==0.001) 11

29 Table 1: Number and Success Rate of Women in Congress by Party, Year Number of Democratic Women Candidates Democratic Women Primary Win Rate Number of Republican Women Candidates Republican Women Primary Win Rate Democratic Women Win Rate Advantage % 81 37% +11.7% % % +4.4% % 50 58% +3.8% % 50 54% +11.2% % % +5.4% % % +4.1% % % +13.3% % % +4.4% % % +19.4% % % +18.9% Data from the Center for Women and Politics (CWAP) 12

30 Figure 2: Female Success in Primaries Across Parties in U.S. House Elections, PARTY DIFFERENCES LEAD TO AN EMERGING PUZZLE The dilemma of the counterstereotypical Republican is clearly a challenge for the Republican Party, particularly today. The urgency motivating the Party s relatively new push for a more diverse constituency has emerged not only due to demographic shifts in the American populace but because the Party has lost considerable ground with non-white and non-male voting blocs in recent decades. The Republican Party, of course, was originally founded as an anti-slavery party, and enjoyed widespread supported from emancipated slaves and their descendants until the Great Depression when Franklin Roosevelt s New Deal policies appealed to African American citizens deeply affected by rampant poverty. In the 1960s, civil rights legislation supported by Democrats (namely President Lyndon Johnson) in conjunction with the anti- 13

31 government rhetoric from leading Republicans (namely Barry Goldwater) precipitated additional migration of African Americans to the Democratic Party. 9 In American elections since that time, African American voters have represented one of the most reliably Democratic voting blocs in the country. The Republican Party s reputation with women has similarly fluctuated over time. The Party s policy vision in the mid-twentieth century included a platform aligned to a large degree with women s interests. Indeed, in the 1950s and early 1960s, Republicans were relatively more favorable to women s rights than were Democrats (Wolbrecht 2002, 238). 10 This edge began to erode in the 1970s, a period during which the parties were not particularly distinct in terms of support for women s rights, before polarizing in the opposite direction starting around Party realignment among women has since manifested itself in a variety of ways ranging from higher rates of Democratic Party affiliation among women (see below) to an increasing prominence of women s rights at Democratic Party conventions since the 1970s to a higher proportion of women in Democratic congressional delegations (Wolbrecht 2000). The partisan gender gap, however, does not parallel the partisan racial gap. While female individuals tend to support the Democratic Party more than men to this day, there is far more partisan parity between the sexes than between the races. As the figures below show, Republican women are not uncommon whereas Republican African Americans remain fairly rare among the citizenry. The evolution of party reputations on minority and women s rights has resulted in the modern political reality that a majority of non-white and non-male candidates affiliate and/or 9 It should be noted that this grossly oversimplifies the situation. Many Southern Democrats in this era, for instance, resisted civil rights legislation, and in fact a higher proportion of Republican legislators in Congress (about 80% in both chambers) support the 1964 Civil Rights Act than did Democrats. However, the reputation of the parties on civil rights have diverged considerably in the years since. 10 For example, for nearly four decades prior to 1980, the Republican Party officially endorsed the Equal Rights Amendment while Democrats opposed it until 1972 (Frum 2000). 14

32 support the Democratic Party, even though the size of the gap is far larger across race than it is across gender. Nevertheless, this reality represents an electoral problem for Republicans in an increasingly non-white American populace. It also represents a puzzle to scholars of cues and voter behavior. Research in political science implies that female or African American Republicans will struggle to achieve electoral success because their demographic status implies ideological liberalism. Yet if the same question is approached from a psychological perspective, an entirely different set of theoretical expectations emerges. Psychological literature on attribution theory and expectancy violations argues that individuals tend to legitimize political objects (candidates, messages, etc.) that fail to conform to expectations. Messages that appear incongruent with the interests of the message provider (e.g. Republican party membership for an African American whose race is not associated with that party) become more powerful because messengers become more authentic (Walster, Aronson, and Abrahams 1966). This phenomenon is known not only to increase the persuasiveness of political messages but also to increase positive affect for counterstereotypical messengers (Bergan 2012; Eagly, Wood, and Chaiken 1978). Hayes (2005), for instance, finds that candidates whom voters perceive to possess counterstereotypical qualities (like a Republican candidate viewed as compassionate or a Democrat who appears to be tough on crime) can increase their appeal to voters. We are thus left with an intriguing puzzle: how is it that being a counterstereotypical Republican candidate is problematic according to political scientists yet ostensibly a beneficial status if we draw from psychology research? One of the key goals of this project is to help reconcile these two literatures and determine when and how counterstereotypical status can be an electoral asset or liability in primary elections. 15

33 1.3.1 Can Democrats be Demographically Counterstereotypical? Absent thus far is a discussion of counterstereotypical Democratic candidates. Do African American or female Democrats face similar struggles as their Republican counterparts? My answer is a qualified no. By definition, one reason the Republican Party has struggled to attract non-white and non-male voters to its ranks is because these citizens tend to support the Democratic Party. Most African American voters report greater affect for the Democratic Party. Figure 3 summarizes this trend in recent years. Surges in African American registration and turnout in 2008 for the first African American presidential candidate a Democrat further affirm this effect. Since the Democratic Party has in recent decades been increasingly associated by all voters as being the party for African Americans, it follows that most African American political candidates run as Democrats. Simply put, being a Democratic African American voter or politician is not unexpected. 16

34 Men Women African Americans Data from Pew Research Democrats (%) Republicans (%) Independents (%) Figure 3: Party Identification by Demographic Group, A similar pattern of support for the Democratic Party exists among women, although the gap in party preference is considerably smaller. As Figure 3 demonstrates, in % of women aligned with the Democratic Party compared to only 24% who favor Republicans. Since 1996, the Democratic presidential candidate has enjoyed winning a majority of female votes while the Republican candidate has enjoyed winning more men. 11 For instance, in 2012 Barack Obama (D) won 55% of women s votes to Mitt Romney s (R) 44% while Romney enjoyed a 52% to 45% edge over Obama among male voters. If we look at female candidates in Table 1, we observe that since 1996 only one election season (2010) featured more Republican women running in primary elections than Democratic women. As with African Americans, women have historically supported and run for office as members of the Democratic Party at higher rates than

35 the Republican Party. A female Democrat is not an unusual sight for a voter. If we consider the representation of women in Congress across the parties we see more evidence that being a female Democrat politician is more or less normal while being a female Republican remains the exception. Consider Figures 4 and 5. Since the early 1990s (recall 1992 was The Year of the Woman (Atkeson 2003)) in both the U.S. House and U.S. Senate, the number of women in the Democratic ranks of Congress has grown larger women in on the Republican side in terms both of (1) the total number of women in the party caucus as (2) the proportion of the caucus comprised of women. Whereas in 1991 there were 19 Democratic female Representatives to the Republicans nine, by 2013 that gulf had widened to 58 Democratic women to 19 Republicans. In the Senate, Republican women have never numbered higher than five, while Democratic women have been steadily increasing to over 15 by As a proportion of seats held by a party, the results in the Senate are even more convincing. Republican women have never held more than ten percent of Republican seats, while today women hold just under 30 percent of Democratic seats. As far as representation in Congress goes, the last twenty years have seen women become increasingly popular in the Democratic Party while their presence in the Republican Party remains stagnant. Put differently, female Republican politicians are counterstereotypical and female Democratic politicians are not. 18

36 Figure 4: Women's Representation in U.S. House by Party, Figure 5: Women's Representation in U.S. Senate by Party, It should be noted that this discussion is not meant to imply (1) that women and African Americans have achieved equality descriptively or substantively even within the Democratic 19

37 Party; or (2) that counterstereotypical politicians cannot exist within the Democratic Party. Instead, I suggest that because the party has successfully appealed to racial minorities and women, and since most female/minority candidates are Democrats, the notion of a demographically counterstereotypical Democrat is an unlikely one insofar as there is no demographic status a Democratic politician can hold that would strike voters as unexpected. In other words, anyone looks like a plausible Democrat. While the 2008 Democratic primaries eventually became a high-profile race between a female and an African American, the original field of candidates also included five white males and one male of mixed racial descent. 12 All of this said, Democrats can just as easily be counterstereotypical in non-demographic domains. Political parties have reputations among the voters (Hayes 2005), and a Democrat with traits or attributes inconsistent with her party s reputation may be classified as counterstereotypical. Similarly, a Democrat maintaining membership in certain groups or organizations Evangelical Christians or the Tea Party, for instance would strike individuals as counterstereotypical just as an African American committed to the Republican Party is. This project, however, focuses on the use of demographic cues when evaluating primary election candidates in a general sense. In this way, some Republican politicians can be counterstereotypical, but Democratic politicians cannot. 12 These include Senator Joe Biden, Senator Chris Dodd, Former Senator John Edwards, Former Senator Mike Gravel, Representative Dennis Kucinich, and Governor Bill Richardson, whose heritage includes elements of Caucasian and Hispanic race. 20

38 1.4 SUMMARY AND NEXT STEPS We have now arrived at the crux of the challenge I will address in the following chapters. Republican Party leaders have recognized the Party s limited ability to appeal to new demographic blocs of voters that historically it has not relied on to achieve electoral success. These same leaders acknowledge that they must do something to respond to this development, for the proportion of white, male individuals within the general electorate bread and butter voters for the Party is decreasing. Republican strategists are encouraging the Party to nominate and elect more women and racial minorities, surmising that descriptive representation signifies one avenue for the party to make inroads with female and minority voters. I suggest, however, that there is reason to be wary of this strategy, and, paradoxically, the challenge to Republican Party success in the future actually lies with Republican voters. Political scientists have well-established that voters infer a great deal about politicians on the basis of demographic features like race and gender. In general, this sort of inference may actually be detrimental to the electoral prospects of female or African American Republicans, namely because their gender and race, respectively, tend to be associated with the Democratic Party and a more liberal ideology. While these traits may be beneficial in a general election where those same associations may allow female or African American Republicans to appeal to more centrist or liberal-leaning voters, these candidates will never make it on the ballot in November without winning primary elections in the spring. Given that primary elections represent the first hurdle in securing a party s nomination, and coupled with the questionable remedy laid out by Republican Party leaders, I submit that a more dedicated scholarly effort must be aimed at understanding how race and gender work in primary elections. Above I identified a series of conditions unique to the primary process, and 21

39 the scholarly literature thus far does not sufficiently explore how demographic cues affect decision-making in this context. We must fundamentally focus on both voters and candidates and develop a theoretical basis for examining the consequences good and bad for different types of political candidates in party primaries. Can (negative) stereotypes be neutralized? Can counterstereotypical candidates benefit from violating expectations? What should a theory concerning the use of race and gender stereotypes in primaries cover? In the next chapter I developed a two-pronged approach that focuses on voters and candidates, respectively. The first part of the theory draws on spatial logic and heuristics literature to argue that there is an additive effect of partisanship and demographic status when it comes to evaluating primary candidates. I argue that since voters both share a partisan affiliation with all primary candidates, and therefore maintain have a generalized sense of candidates ideologies and policy views, they may use race and gender more precisely to locate, or sort, these candidates within a party s ideological space. From here, voters own ideological dispositions should influence their evaluations of their primary options. In other words, since a voter can select from several candidates representing the voter s party, race and gender will help her decide which primary choice is most aligned with her own ideological disposition, and she is expected to express more favorability towards candidates she perceives to be most ideologically like herself. The second part of the theory approaches the issue from the candidates perspective. Above I briefly established that some primary candidates are counterstereotypical, not fitting voters preconceived notions about their policy and ideological views. The classic dilemma of the counterstereotypical Republican argues that these candidates generate confusion and, subsequently, voters will withdraw support from them. How is it, then, that these candidates can 22

40 succeed in Republican primary elections? The answer, I suggest, comes in the form of policy information. When demographic information conflicts with partisan status, voters may struggle to accurately infer about primary candidates. A candidate can remedy this by supplying ideological information that allows voters to move away from conflicting categorical information and focus on more precise and relevant policy information. This scale-tipping information acts to resolve the cognitive conflict that results from categorical cues implying opposite ideologically dispositions. From here, I posit that it may even be possible for counterstereotypical candidates to activate positive affect via expectancy violations given that they have established themselves within ideological subgroups within parties Project Outline The rest of the chapter is organized as follows: Chapter Two reviews the relevant literature that I seek to expand with this work. I then develop a theory of primary decision-making in which I integrate research on demographic cues as ideological stereotypes with basic spatial mapping processes to predict when certain types of primary voters will prefer certain types of primary candidates. I then go on to discuss the concept of individuating information and develop a new variant of it that emerges when other, categorical pieces of information are in conflict. Theoretical and empirical expectations for both parts of the theory are presented. Chapter Three provides details of two survey-experiments I conducted to test the hypotheses. Chapters Four through Six provide empirical analysis of the theory and its implications. Chapter Four presents results from the low-information analysis when only categorical information is provided to respondents. Chapters Five and Six analyze results from the second study in which individuating policy information is introduced. The first of these chapters 23

41 focuses exclusively on how this information affects perceptions of candidates as they pertain to ideology, and the second emphasize related attitudes about politicians as potential party candidates. Chapter Seven closes the project, summarizing the results of the experiments, discussing their limitations, and discussing the implications for gender- and race-based politics moving forward. Finally, since this project is by no means the last word on race and gender in primary elections (indeed, it is one of the first), I also spend time in the final chapter discussing how the results of this work will contribute to additional study on this topic. 24

42 2.0 CHAPTER TWO: THEORY Conventional accounts of the application of demographic stereotypes in rendering evaluations of politicians conclude that counterstereotypical African American and female candidates in the Republican Party are trapped in an information dilemma for voters attempting to render judgments about them. These candidates are members of the conservative political party yet individually are presumed by voters to be more liberal than white males (Huddy and Terkildsen 1993a; Koch 2000; Mcdermott 1997). 13 Consequently, they struggle when running as Republicans (King and Matland 2003). While these concerns are not unfounded, the bases for them are not adequately established. Almost all work on the role of race and gender cues has been studied in either a non-partisan context (e.g. Jones 2014; Mcdermott 1998) or a general election context where race and gender cues are drown out by partisanship (e.g. Dolan 2004; Hayes 2011; Koch 2000, 2002). Intra-party analysis of demographic cues in primaries has typically been restricted to specific election cycles (e.g. Jackman and Vavreck 2010). As established in Chapter 1, the unique nature of primary electoral politics suggests that generalized claims about the role of race and gender in voter decision-making do not necessarily apply in the same way they do in the fall. In the following pages I propose an alternative theoretical formulation of how these cues are applied in intra-party contests. The chapter is 13 Koch (2002) finds this bias to occur even when controlling for candidate s actual ideology. 25

43 roughly divided into three parts: first, I briefly chronicle our extant knowledge of racial, gender, and partisan cues in American elections, emphasizing their role as ideological signals in the candidate evaluation process. I then integrate these stereotypes into basic spatial logic, arguing that primary voters use race and gender to sort primary candidates and identify their most preferred candidate. Finally, I explore the role of political information in primaries, formulating a model of scale-tipping information that suggests counterstereotypical candidates can provide policy messages that compel voters to focus on the ideological substance of the candidate s policy views and not the conflict generated by categorical cues. This, in turn, allows counterstereotypical candidates to level the electoral playing field, and perhaps even reverse it., Ultimately, this project seeks to provide some answers to two central questions that have been heretofore un- or under-appreciated in the cue literature: the first concerns the nature of subgroups within parties: how do ideological moderates behave differently than ideological extremists? Second, how does policy information condition the impact of other information cues that have historically been studied in isolation? 2.1 USING (EASY) INFORMATION: RACE, GENDER, AND IDEOLOGICAL INFERENCE There is little debate in modern political science that voters are by and large cognitive misers, interested in minimizing mental output in the course of making political decisions. As Fiske and Neuberg (1990, 14) put it, we are exposed to so much information that we must in some matter simplify our social environment for reasons of cognitive economy, we categorize others as members of particular groups groups about which we often have a great deal of [ ] 26

44 stereotypic knowledge. In an effort to make reasonably informed decisions without incurring the costs required to obtain the information required to do so, voters turn to cues about candidates. A cue is a simple, easily-obtained piece of information about a candidate which facilitates inference about the candidate as a whole. One of the best documented of political phenomena is the tendency for voters to use cues to aid in decision-making in low-information contexts (e.g. Lau and Redlawsk 2001); indeed, as information decreases, reliance on cues increases (Matson and Fine 2006). Voters often use cues to place political candidates into categories (Fiske and Taylor 1991). Once in these categories (race, party, gender, etc.), voters can assign attributes to particular candidates based on characteristics they associate with the category as a whole. Race and gender represent two of the most well-developed cue literatures in social science and here I briefly discuss each in turn. Research on racial cues has affirmed that they activate negative feelings towards African Americans among white voters and can stimulate anti-african American policy preferences across a variety of contexts, from general policy issues like support for disaster assistance (Iyengar and Hahn 2007) to race-related policy domains like welfare, affirmative action, and immigration (Brader, Valentino, and Suhay 2008; Federico 2004; Hutchings and Valentino 2004; Valentino 2002) to elections (Citrin, Green, and Sears 1990; Jackman and Vavreck 2010; Mendelberg 2001; Peffley and Hurwitz 2005, 2007; Tesler and Sears 2010). White voters penalize African American candidates in their evaluations while rating white candidates more positively (Berinsky et al. 2011; Best and Williams 1990; Terkildsen 1993). 14 Indeed, as race becomes more salient among white voters, they are increasingly less likely to support African 14 Subsequent generations of research have probed such cues more deeply. For instance scholars have even found that skin complexion matters: the lighter a black candidate s skin, the more acceptable he is to white voters (Maddox and Gray 2002; Weaver 2012). 27

45 American candidates (Schaffner 2009). 15 These patterns contribute to African American candidates only rarely elected achieving electoral success outside of majority-minority districts. 16 Gender cues have historically shared some common characteristics and consequences as racial cues. Both cues remain capable of activating negative feelings towards the group depicted. 17 Armstrong and McAdams (2009), for instance, find that individuals are more likely to rate a message favorably if it is attributed to a man rather than a women. However, electorally, women have achieved greater electoral parity with men than have racial minorities; for instance, women do not appear to suffer from systematic bias at the polls (Cook 1998; Duerst-Lahti 1998; Fox 2000; Smith and Fox 2001). Nevertheless, electoral equality has not been achieved; women, for example, are far more susceptible to primary competition than men (Lawless and Pearson 2008). The figures from the last chapter also highlight the persistent gulf in descriptive representation between men and women in Congress, especially among Republicans. Fundamentally, however, why do racial and gender cues activate attitudes towards these politicians? Part of the answer is found in classic models of racism and sexism that stimulate biases against non-white and non-male candidates (e.g. Githens and Prestage 1977; Kirkpatrick 1974). Yet there is additional explanation, as well: given their desire to simplify the political 15 Some scholars have argued against this trend, suggesting that white voter discrimination against blacks does not appear in election data (e.g. Highton 2004). 16 (Canon 1999, 10) points out that in the 6667 House elections in white majority districts between 1966 and 1996 (including special elections), only 35 (0.52%) were won by blacks. 17 Women and African Americans share other group characteristics, as well; for instance, both are known to be uniquely capable of influencing behaviors stemming from shared group status. Members of underrepresented groups more sensitive towards issues impacting their group, and consequently are particularly more likely to support them. Women have a stronger preference for same-sex representation than men (Dolan 2008). Atkeson (2003) finds that women are more likely to participate in elections when female candidates are running. Female candidates also enjoy marginally greater support among voters of their own gender (Brians 2005; Dolan 2004). 28

46 environment when pressed to make electoral decisions, voters tend to impose on political candidates ideas and attitudes they associate with groups of which candidates are a part. Party labels, for instance, help voters make an informed choice by associating specific, unknown candidates with known (generalized) policy views (Aldrich 1995; Downs 1957; Mondak 1993; Rahn 1993; Schaffner and Streb 2002). Voters, consequently, tend to view candidates through the lens of party reputations (Hayes 2005). Race and gender activate similar associations. 18 Specifically, African Americans and women are viewed differently than white men on three dimensions of political attributes: (1) traits, (2) issues, and (3) ideology. Men, for example, are perceived as tough, aggressive, self-confident, and assertive while women are more often viewed as compassionate, gentle, kind, and caring (Huddy and Terkildsen 1993a; 1993b). Assessments of candidate traits have been shown to activate both positive and negative attitudes about the candidates themselves (Kahn 1996; Lawless 2004; Sanbonmatsu 2002). Furthermore, trait perceptions have been found to influence vote choice, both in presidential (Bishin, Stevens, and Wilson 2006; Hayes 2009) and subnational elections (Druckman 2004; Fridkin and Kenney 2009; Hayes 2010). Individuals are also known to associate candidates of different demographic groups with particular policy interests and competency in specific issue areas. African American and female candidates, for instance, are significantly more likely to be perceived as interested in minority issues than are white men (Mcdermott 1998). Sigelman et al. (1995) find that minority status boosts the perception that moderate or conservative candidates would be compassionate towards 18 For this reason, I justify treating women and African Americans as theoretically similar in this particular context, for both groups are associated with the same underlying political ideology. As I discuss in the closing chapter, however, this does not mean that these groups or any others can always be presumed to generate similar feelings within individuals. 29

47 disadvantaged groups. More generally, men are perceived as better equipped to oversee policy in certain domains national security, for instance while women are presumed to be more competent in areas like education and healthcare (Alexander and Andersen 1993; Kahn 1996; Lawless 2004). Finally, and most importantly for my purposes, impressions of politicians ideology are strongly influenced by their demographic status. 19 Voters perceive female political candidates to be significantly more liberal than male candidates overall (Koch 2000, 2002; Sigelman et al. 1995; Williams 1990). Similar patterns exist even when politicians are members of the same party. 20 King and Matland (2003) find that voters tend to believe the term conservative is more fitting for a male Republican than a female Republican. Dolan (2004) adds that female Republican incumbents are perceived to be more liberal than male Republican incumbents. Jones (2014) finds that even when political candidates present policy views which should reduce reliance on demographic cues voters still perceive non-white politicians as more ideologically liberal than white males, regardless of the policy view they profess. In short, two main points have been summarized in this section. First, individuals are highly dependent on heuristic cues in the course of making political decisions. Second, individuals hold ideological stereotypes about politicians on the basis of both partisan affiliation 19 Not all demographic research finds significant differences in impression formation as a function of differences between demographic groups. Brooks (2011), for instance, finds that voters do not evaluate men and women differently based on displays of emotions. Others fail to find a relationship between a voter s ideology and the race of a politician she is evaluating (e.g. Colleau et al. 1990; Weaver 2012). Most studies, however, find significant effects. 20 In the abstract, demographic-based assumptions about candidate ideology are not necessarily unjustified. As the first chapter illustrated, most African American and female voters do report greater affect for the Democratic Party. Tate (1993, 1) even observes that in the 1992 elections [blacks] voted overwhelmingly Democrats, as they had in the last seven presidential elections [ ] [This] despite the fact that the Democratic presidential nominee, Bill Clinton, had done little to earn their votes. Surges in black registration and turnout in 2008 for the first black presidential candidate a Democrat further affirm this effect. Women also support the Democratic Party at higher levels than white men do, though the difference is not as robust as it is for racial minorities. 30

48 and demographic information like race and gender, where white male candidates are perceived to be more ideologically conservative than all other demographic groups. Thus it becomes all the more likely that when partisanship is constant across candidates, primary voters may use demographic information as ideological information. A priori, given the well-documented association between the Democratic Party/ideological liberalism and women and racial minorities, primary voters who encounter non-white candidates should use this information to modify their views of those candidates ideological positions within a pool of party candidates. From here, voters perceived ideological congruence with different candidates should condition the degree to which candidate evaluations are favorable. In the next sections, I argue that demographic cues as well as other candidate attributes namely, policy messages signal important information that aid in the decision-making process by helping voters ideologically sort candidates within their own parties. 2.2 HOW VOTERS EVALUATE PRIMARY CANDIDATES: ASSUMPTIONS UNDERLYING INFORMATION USE A model of information-based sorting and selection begins with a several critical assumptions concerning voters goals and objectives when using cues in primary elections. First, voters are presumed to know and access generalized stereotypes about political and demographic groups. While many voters are not highly informed about politics overall, the cue literature discussed above provides a reasonable basis to make this claim. A second assumption is more specific to intra-party contexts. Individuals in a primary setting are expected to use information to sort a group of like-minded candidates in a way that allows the individual to identify which party 31

49 choice is best suited to represent his or her interests. Unlike in general elections where copartisanship between voter X and candidate Y may be sufficient for that X to support Y, in primaries ideology should be more central to the act of identifying a most preferred candidate. 21 Utilizing a basic spatial framework (e.g. Downs 1957), we should expect that an individual will generally prefer candidates she presumes to be ideologically proximate to herself more than she will prefer candidates she perceives to be more ideologically distant, even when multiple candidates share partisan status with her. Consequently, both general demographic and specific information cues are presumed to be utilized by individuals to sort candidates ideologically for the purposes of identifying ideological congruence with various candidate options. I also assume voters across both parties utilize demographic cues in a similar fashion as they concern ideological inference. White male politicians have historically been the most common (and even the exclusive) demographic type across both parties in American politics. As a result, they can plausibly subscribe to any ideology, but in the absence of ideological or partisan information voters cannot ideologically distinguish one white male candidate from other. Yet relative to white male candidates, female and African American politicians may be perceived to be more liberal on the basis of demographic status alone. In other words, in the absence of information signaling candidates ideological disposition, voters should not be able to infer and ideological difference between two men, but will be more likely to infer an ideological difference between a male and female, or white male and black male. 21 Indeed, in a general election, even if race or gender have the power to alter impressions about candidates to some degree, it is unlikely voters would defect from their own party to support the other. Scholars have wellestablished that individuals tend to prefer members of social in-groups to people with whom they do not share some status (Leonie Huddy 2001; Tajfel and Turner 1979). This is the main reason extant literature on the intersection of gender and partisanship (e.g. Hayes 2011; Sanbonmatsu and Dolan 2009) is theoretically inadequate for specifying the role of demographic cues in primary elections. 32

50 Given the universality of liberal stereotypes associated with both female politicians and politicians of color, for instance, I assume that knowledge that a candidate is female will compel voters in both parties and of different ideological groups to all presume that candidate to be somewhat more liberal than they would, for instance, presume a white male candidate to be. By the same token, voters should generally perceive any white male candidate to be somewhat more conservative than any African American male candidate. This assumption, however, should not be taken to mean that the effects of demographic information on ideological perceptions are necessarily equivalent across different types of candidates. Here I am merely positing that the direction of any shift in ideological perception on the basis of race or gender will be the same across voters. This project also assumes that individuals will typically give more weight to information from a surprising or unconventional source than information from a stereotypical or expected source. When candidate information is consistent (for instance, when only one piece of information is available or when multiple pieces of information like race and party are ideologically aligned), it represents a sort of cognitive equilibrium in voters minds. All the available information, after all, makes sense. On the other hand, when information is introduced that is inconsistent as when a female candidate affiliates with the Republican Party this equilibrium is disrupted and the voters are much more apt to modify their perceptions of the candidates. A fifth and final assumption concerns the nature of information available to individuals as they effort to evaluate political figures. We may loosely characterize partisan and demographic cues as easy, or categorical, for they connote broad associations with ideological and other types of stereotypes. Easy cues can be quite useful in generalizing about particular 33

51 objects, and they have the added benefit of high accessibility voters need not invest a great deal of energy in acquiring easy information. For instance, a voter watching a primary debate would not have to listen to a word uttered by any candidate but could still easily distinguish among candidates on the basis of race and gender. Yet despite its accessibility, easy information can be detrimental in that it is often imprecise. Specific objects may not always fit into the categories with which they are associated. Indeed, counterstereotypical Republican politicians exemplify this very problem. The challenges posed by easy information can sometimes be rectified by more complex, candidate-specific information. This individuating, or hard, information refers to knowledge that is more precise than the general stereotypes that are associated with easy information. Since individuating information is more exact, voters should find it more relevant that easy, categorical information and therefore give it more weight in the evaluative process. Individuating information, then, while more costly to obtain, should also be more influential in voters minds. Below I argue that when counterstereotypical candidates can resolve cue conflict through ideological signals embedded in individuating policy messages, they can overcome the challenges posed by ideological associations with race and gender in Republican primaries. In summary, the theory outlined in these pages relies on a set of assumptions about knowledge individuals have as well as how they use knowledge they are presented. I implicitly assume voters have a set of attitudes and beliefs related to (1) partisan, (2) racial, and (3) gender cues. I also anticipate that voters will use information easy and hard to ideologically sort primary candidates, and will generally prefer the most ideologically proximate option as their choice for their party s nomination. Furthermore, the more information is (1) unexpected and/or (2) specific, the more relevant it will be as voters attempt to evaluate primary candidates. This, 34

52 as I argue below, allows counterstereotypical candidates to level the electoral playing field, and perhaps even reverse it. 2.3 SORTING AND SELECTING WITH EASY INFORMATION The process of candidate sorting using cues often entails utilizing multiple pieces of information to subdivide, or subtype, voters into different groups. Subtyping theory (Deaux et al. 1985; Devine and Baker 1991) posits that individuals are known to sort objects within global categories like men. This sorting creates subtypes, or multiple groups which are similar on some dimension but different on another. In party primaries, white male Democrats may constitute one subtype of Democrats while white female Democrats constitute another. Scholars have found strong evidence of subtyping effects individuals perceive differences across subgroups even when all members of those groups share a common global status. Recent work by Schneider and Bos (2011) finds that African American politicians as a group are perceived differently than African Americans as a whole. African American politicians, for instance, are considered more ambitious, confident, and educated than African American citizens. Indeed, African American politicians in some ways share more group overlap with African American professionals and politicians in general than they do a simple racial group. Applying the same logic to female politicians, Schneider and Bos (2014) again find that female politicians are not perceived to possess the same characteristics that voters ascribe to females in general. As noted in the previous chapter, when primary voters are evaluating political candidates within their own parties, the need to distinguish among electoral choices is far more urgent than it is in general elections. Demographic information provides a relatively costless, if not crude, 35

53 method for such discrimination. While a primary voter may have general ideological expectations about a co-partisan, they may amend those expectations based on demographic information that accompanies the party cue. In general, across both parties, we should expect female and African American candidates to be seen as more ideologically liberal than their white male counterparts. Formally: H1: Democratic individuals will perceive female and African American Democratic candidates to be more ideologically liberal than white male Democratic candidates H2: Republican individuals will perceive female and African American Republican candidates to be more ideologically liberal than white male Republican candidates Establishing ideological distinctions, however, is not sufficient for determining whether demographic cues lead different types of voters to conclude that some candidates are more ideologically proximate than others. Thus another ideological measure must be considered. Just as I expect voters to use demographic cues to establish a candidate s ideology, I also expect that demographic cues help establish ideological congruence between a primary voter and a particular candidate. Formally: H3a: The more ideologically liberal a Democrat, the larger the perceived ideological gap between herself and a white male Democratic candidates H3b: The more ideologically conservative a Democrat, the larger the perceived ideological gap between herself and a female and African American Democratic candidates H3c: The more ideologically liberal a Democrat, the smaller the perceived ideological gap between herself and female and African American Democratic candidates 36

54 H3d: The more ideologically conservative a Democrat, the smaller the perceived ideological gap between herself and a white male Democratic candidates H4a: The more ideologically conservative a Republican, the larger the perceived ideological gap between herself and female or African American Republican candidates H4b: The more ideologically liberal a Republican, the smaller the perceived ideological gap between herself female or African American Republican candidates H4c: The more ideologically conservative a Republican, the larger the perceived ideological gap between herself and female or African American Republican candidates H4d: The more ideologically liberal a Republican, the smaller the perceived ideological gap between herself and female or African American Republican candidates Finally, if, as I have argued, primary voters (1) use partisan and demographic cues to subtype candidates, and if voters (2) prefer more ideologically proximate candidates, different types of voters should prefer (e.g. express support for, perceive to be a good representative) different (demographic) types of primary candidates. In Democratic primaries, the median voter is more liberal than is the median voter in a general election. As a result, it follows that Democrats will perceive female or African American Democrats as more ideologically liberal than white male Democrats despite the shared partisan status. Thus more ideologically liberal Democrats should perceive female or African American Democrats as more ideologically appropriate choices than white male Democratic candidates. For Republicans, the opposite occurs. Since female or African American Republicans will be perceived to be to the ideological left of the median primary voter, more ideologically moderate Republicans should view these counterstereotypical politicians as more appropriate representatives than will more ideologically 37

55 conservative Republicans. This discussion motivates final set of hypotheses is justified. Formally, H5a: The more liberal a Democrat s ideology, the more support she will express for female or African American Democratic candidates H5b: The more liberal a Democrat s ideology, the less support she will express for white male Democratic candidates H5c: The more liberal a Democrat s ideology, the more likely she will be to perceive female or African American Democratic candidates as good representatives H5d: The more liberal a Democrat s ideology, the less likely she will be to perceive white male Democratic candidates as good representatives H6a: The more conservative a Republican s ideology, the more support she will express for white male Republican candidates H6b: The more conservative a Republican s ideology, the less support she will express for female or African American Republican candidates H6c: The more conservative a Republicans ideology, the more likely she will be to perceive white male Republican candidates as good representatives H6d: The more conservative a Republican s ideology, the less likely she will be to perceive female or African American candidates as good representatives In summary, the first set of empirical expectations is drawn from a model of spatial reasoning that integrates demographic stereotyping and basic subtyping theory with Downsian logic to primary elections. Voters are expected to use race and gender to sort political candidates within political parties and form preferences based on where those candidates lie on an ideological spectrum relative to their own. In this way, different ideological voters within parties are expected to form different candidate preferences. Yet useful as this may be as a starting point, this model, with its emphasis on easy, categorical information, represents an incomplete 38

56 treatment of candidates stereotypical and counterstereotypical in primary elections. First, I am for the moment assuming that other information about candidates (e.g. policy views) is unavailable (and therefore constant across candidates). This is not entirely unreasonable after all, earlier it was established that voters are known to be generally uninformed about specific candidates and are highly dependent on heuristic cues. However, a greater challenge remains, at least for Republicans: the dilemma of the counterstereotypical Republican is based on the fact that counterstereotypical candidates generate cognitive conflict. This may lead moderate Republican voters to be ambivalent among Republican candidates of any race or gender. Without more information about candidates, these moderate voters the very ones counterstereotypical candidates should target in primaries may not be any more inclined to support African American/female Republicans over conventional white male candidates. I must therefore consider more deliberately how voters process conflicting information and develop a model predicting the circumstances under which counterstereotypical candidates can overcome the (negative) stereotypes with which they are associated. 2.4 LEVELING THE PLAYING FIELD: INDIVIDUATING IDEOLOGICAL INFORMATION Voters use multiple categorical information cues to sort and evaluate primary candidates. Above I argued this this is accomplished by utilizing ideological signals embedded in racial and demographic cues and using them to more precisely locate primary candidates along the ideological spectrum within political parties. Focusing exclusively on easy, categorical 39

57 information, however, does not resolve the cue conflict problem voters encounter in Republican primary elections. Given the prevailing liberal stereotypes associated with women and racial minorities, Republican voters are not likely to prefer counterstereotypical candidates to conventional ones when demographic and partisan status are the only pieces of information available to them. How if at all can these candidates be electorally viable in primary elections? The answer to this question requires us to move beyond global cue categories like race or party and focus on the role of individuating, candidate-specific knowledge in a primary context. I submit that candidates can use this information to establish their ideological identities within parties and, once this has occurred, it becomes possible a candidate s status as counterstereotypical to stimulate positive affect stemming from violating information expectations. While scholars tend to emphasize categorical-based cue inference in candidate evaluation (Bianco 1998; Conover and Feldman 1989; Lodge, Steenbergen, and Brau 1995), approach is not sufficient for decision-making in party primaries. An alternative evaluative route is based on an individuating approach in which candidate perception is a function not of categorical stereotypies (e.g. Democrats are pro-choice) but instead on specific information obtained about a particular candidate (e.g. she is pro-choice). 22 In this section, I apply this processing to the primary context. Since conflicting cues only exist in Republican primaries, I look only at Republican candidates and voters for the moment. 22 Much of the early literature on individuating information did not consider demographic conditions. Riggle et al. (1992), for example, study the impact candidate attractiveness, party membership, and voting record (2x2x2 design). They find, among other things, that candidate attractiveness has a strong impact on overall candidate evaluations when no other information is provided about a candidate, but this effect disappears when individuating partisan or ideological information is provided to voters. Budesheim and DePaola (1994) employ a 2x2 design in which physical attractiveness versus personality information are contrasted with either favorable or unfavorable image information. Unlike Riggle et al., they find that physical appearance influence candidate evaluation even when individuating personality information is provided. 40

58 The key to individuating information within Republican primaries is that it should work differently for African American and female candidates than it does for white male candidates. African American and female candidates, we have established, send mixed categorical signals (Hayes 2011; Sanbonmatsu and Dolan 2009), which represent a cognitive challenge for Republican primary voters: which information dominates, the partisan label or the demographic cue? Mcdermott (1997), summarizes the problem well: because a Republican woman provides voters with two competing [categorical] cues, [ ] voters may not know which cue to give more weight (278). This conundrum implies that these candidates must find a way to signal their ideology to voters. One way candidates may do this is to provide specific information, or individuating information, to redirect voters attention away from the tension between conflicting group categories (partisanship and demographic status). In general, Individuating information has been shown to achieve one of two things: first, it may lead individuals to discard or discount categorical cues, as they prefer instead to use more precise individuating information to form impressions. Crawford et al. (2011), for instance, finds that voters increasingly draw on individuating information about political candidates as it becomes more relevant to the task of rendering judgment. 23 As the amount of information to which people are exposed increases, the less likely they are to rely on, for instance, gender cues in impression formation (Banducci, Everitt, and Gidengil 2002; Chang and Hitchon 2004; Pratto and Bargh 1991). Arceneaux (2008) finds that voters focus on candidates individuating policy views rather than their partisan status when candidates present conflicting information (e.g. a conservative Democrat). More 23 For instance, if a voter is asked about a candidate s view on stem-cell research, individuating information about the candidate s views on abortion are more likely to be brought to bear than are the candidate s views about tax policy. 41

59 recently, Boudreau and MacKenzie (2014) show that voters do not blindly follow partisan cues specific policy information can mediate public opinion even when coupled with party labels. In short, as individuating information enters the fray, global categorical cues become less relevant. Alternatively, individuating information can also serve to help voters subtype, using a similar process as was introduced in the last section. Above, subtyping was conceived as categorical information race and gender modifying other categorical information party labels thereby allowing voters to ideologically sort candidates within parties and helping them identify the differences that allow them to select one primary candidate over others. That process, however, results in cue conflict for one group of candidates counterstereotypical Republicans. As a result, unlike most applications of individuating information, the mere act of subtyping a group of Republican political candidates does not necessary generate more clarity for Republican voters. It may actually complicate the primary picture. To move from complicating to clarifying candidate choice in Republican primary elections, I propose a slightly revised variant of individuating information. In complex political environments individuals may be faced with contradictory categorical information. When only categorical information is present, as discussed above, this cue conflict cannot necessarily be resolved. Yet a specific type of individuating information what we may call scale-tipping information may aid this process. Scale-tipping information does not merely (1) serve to replace categorical cues or (2) help classify candidates within categorical groups, but instead provide an ideological signal that has primacy relative to categorical cues that are in conflict. 24 In this way individuals can use scale-tipping information to simplify the evaluation process: when categorical cues are in tension, individuating information allows individuals to bypass the 24 Individuating information more generally can refer to anything specific information that details to voters. Not all such information, though, may help voters tip the scales towards one categorical cue or another. 42

60 cue tie by focusing on more precise, and therefore more relevant, ideological information. Consider an African American candidate running in a Republican primary. By himself, he represents cue conflict, and Republicans may assume he is politically moderate, or struggle to classify him at all out due to uncertainty. However, if he establishes himself as politically conservative through individuating policy information, individuals can focus on these specific ideological credentials and dismiss the cue conflict. This accomplishes two important objectives: first, uncertainty is reduced, and voters will be less likely to reject the candidate because they struggle to infer his ideological proclivities. Second, individuating information provides a clear ideological signal that allows Republican voters to clearly sort this Republican along with others. This scale-tipping variant of individuating information, by definition, will apply to African American or female Republican candidates, but it will not typically apply to conventional white male candidates. It is counterstereotypical candidates, after all, that present conflicting cues, and thus these candidates have more to lose or gain from individuating information than others candidates do. White male Republicans do not represent conflict, and therefore individuating policy information, while not irrelevant to voters, should be somewhat less relevant when evaluating conventional candidates relative to evaluating female or African American politicians. A critical step in this process concerns the ideological direction implied by individuating information. If an African American or female Republican provides individuating information that implies a more conservative worldview (say, an anti-tax pledge), voters will reject the liberal signal contained in the demographic cue and base their evaluations of the candidates on the ideological substance of the tax pledge. On the other hand, if the candidate is described as 43

61 holding more moderate or liberal policy views (say, supporting a capital gains tax), this sends a different ideological signal and the conservative ideological cue associated with the Republican Party label is largely dismissed in lieu of specific information implying ideological moderation. 25 In short, the concept of scale-tipping information is simple but central to the electoral viability of African American, female, or other Republican candidates that present conflicting partisan and demographic cues: I propose that categorical cue conflict can be resolved by individuating information that compels voters to attribute to candidates the ideology corresponding with their policy message rather than the ideology embedded within conflicting categorical cues. In this way I expect that female and African American Republicans, given that they provide particular individuating information, will not struggle to earn support relative to stereotypical white male candidates. In other words, when stereotypical and counterstereotypical candidates are similar in terms of their policy preferences, the role of demographic cues should play little role the impression formation process. As a result, these counterstereotypical candidates are in a position to, in effect, level an electoral playing field that is conventionally assumed to work against them. Empirically, this discussion establishes several baseline hypotheses. First, we should expect moderate Republicans to prefer ideologically moderate candidates to more conservative ones regardless of candidates demographic status. We should similarly expect conservative 25 This reasoning does not represent the first effort to unpack the individuating puzzle in a political context. Leonie Huddy and Capelos (2002), for instance, study the interaction between partisan labels and candidate traits. The authors rely on a parallel processing model (Kunda and Thagard 1996) that argues individuals may employ multiple stereotypes simultaneously. They find that when individuating trait information is provided, all gender effects disappear, and the only significant determinant of respondents perceptions of candidate ideology is party labels. Hayes (2011) similarly finds that partisan cues dominate demographic ones. Useful as these studies are, however, they do not speak theoretically to primaries where partisan affiliation is the same for all candidates. Yet, as I have argued throughout, this is precisely the electoral environment in which individuating information is most critical for voters to distinguish among candidates. 44

62 Republicans to prefer ideologically conservative candidates to more ideologically moderate ones. Formally: H7: When political candidates provide the same individuating policy information, race and gender will not impact ideological perceptions of candidates H8a: The more ideologically conservative a Republican primary voter, the more she will perceive as a conservative Republican primary candidate to be ideologically congruent H8b: The more ideologically conservative a Republican primary voter, the less she will perceive as a moderate Republican primary candidate to be ideologically congruent H8c: The more ideologically moderate a Republican primary voter, the less she will perceive as a conservative Republican primary candidate to be ideologically congruent H8d: The more ideologically moderate a Republican primary voter, the more she will perceive as a moderate Republican primary candidate to be ideologically congruent 2.5 EXPECTANCY VIOLATIONS AND POSITIVE AFFECT The theoretical discussion thus far submits that individuating information helps candidates signal an ideology that can clarify for voters the true ideological nature of the candidate. This model of candidate evaluation is fundamentally an informational one, where various cues provide ideological signals that help voters sort and select among primary candidates. Individuating policy information can help neutralize the potentially negative electoral consequences of a candidate s stereotypical status: since uncertainty is an electoral liability, and individuating 45

63 policy information redirects voters attention away from conflicting cues that cause it, this information can help level the playing field for unconventional Republican candidates. Even if voters focus on individuating information while largely discarding categorical information, however, this does not preclude entirely the potential for race and gender to condition evaluations in an affective sense. After all, voters do not simply forget the race, gender, or party of any given candidate they evaluate, even in the presence of individuating information. Instead, we should expect those cues are relegated to secondary status as voters focus on the more precise policy information. This creates a curious condition where female or African American Republicans can close the electoral gap between themselves and white male Republican candidates via individuating information, but their counterstereotypical status remains accessible to voters. When this happens, a candidate s status as counterstereotypical may ultimately be an electoral asset in an affective sense. As noted above, individuals find conflicting information startling preconceived notions about some object are no longer clearly applicable. When this equilibrium is disrupted, individuals are much more prone to update their views of that object since initial impressions are questionable given the conflicting information. Yet what kind of updating occurs? My answer draws from psychological literature on attribution theory and expectancy violations, which proposes that individuals will lend more credence and legitimacy to political candidates who do not conform to expectations. When a message s content can be explained by attributes of the source (for instance, self-interest), then the message has little or no capacity to affect the recipient s attitudes because no new information is provided (Cizmar and Layman 2009; Crawford et al. 2011). Unanticipated messages, on the other hand, are more powerful. Since they confound expectations, they compel recipients to focus more on the 46

64 substance of the information as well as render the messenger more authentic (Walster, Aronson, and Abrahams 1966). In a political context, this phenomenon can both increase the persuasiveness of information as well as increase positive affect for the messenger (Bergan 2012; Eagly, Wood, and Chaiken 1978). Bullock (2011) demonstrates that Republicans find arguments more persuasive when Democrats oppose a stereotypical Democratic policy view (expanding healthcare benefits) than when Democrats support it. Hayes (2005) finds that candidates whom voters perceive to possess counterstereotypical qualities (like a Republican candidate viewed as compassionate) can increase their appeal to voters 26. Schneider (forthcoming) finds that male and female candidates that emphasize political policies not associated with their gender (e.g. a male candidate who mentions that education is a priority for him) improve their perceived levels of competency overall. African American and female Republicans, therefore, may not only be able to use individuating policy information to neutralize any detrimental effects of race and gender, but may also generate positive affect via their counterstereotypical status. This discussion implies a final theoretical point in which an affective component is introduction to the baseline information model. If we take as given that (1) cue conflict is resolved, or at least downgraded, when candidates introduce individuating policy information, and that (2) individuals still recognize demographic cues even when individuating policy information is provided, and, finally, that (3) counterstereotypical messages can activate among its recipients positive affect for the messenger, it is plausible that female or African American candidates may not only level the electoral playing field, but may even tip it somewhat in their favor. 26 In the 2000 presidential election, George W. Bush s attempted to bill himself as a compassionate conservative for this very reason. 47

65 Since individuating policy information is expected to exhibit primacy over categorical demographic and partisan information, any positive affect resulting from counterstereotypical status should be secondary to ideological considerations. Thus, any affective benefits should occur principally after an ideological identity has been established. In other words, counterstereotypical and ideologically moderate Republicans may enjoy positive affective gain (relative to ideologically moderate white males) among ideologically moderate Republican voters. At the same time, within the population of ideologically conservative candidates, female and African American Republicans may enjoy more positive affect among conservative Republicans than would equally conservative stereotypical candidate. In other words, given that an ideological identity is established, counterstereotypical status may activate positive affect among voters within that ideological group of voters. Formally: H9a: The more ideologically conservative a Republican primary voter, the more she will support African American or female conservative candidates relative to white male conservative candidates H9b: The more ideologically conservative a Republican primary voter, the more she will perceive African American or female conservative candidates to be good representatives relative to white male conservative candidates H9c: The more ideologically moderate a Republican primary voter, the more she will support African American or female moderate candidates relative to white male moderate candidates H9d: The more ideologically conservative a Republican primary voter, the more she will perceive African American or female moderate candidates to be good representatives relative to white male moderate candidates 48

66 2.5.1 Policy Information and Democratic Party Choice The Democratic Party has not surfaced in the previous discussion on scale-tipping information, and with good reason: the absence of conflicting cues among Democratic candidates in effect means that voters face considerably less uncertainty across different demographic types of candidates. At the same time, there are no counterstereotypical candidates to violate expectations and capitalize electorally as a result. Empirical expectations for Democrats, then, are considerably more simple than they are for Republicans. Race and gender should have limited effect on evaluations of Democratic primary candidates; instead, preferences should be formed solely on the basis of the ideological nature of candidates policy messages. More ideologically moderate Democrats should prefer moderate candidates of any demographic status and more liberal Democrats should prefer the more liberal candidate regardless of demographic status. Or, more formally: H10: Ideologically moderate Democrats should prefer ideologically moderate candidates to ideologically liberal candidates, but will not support one demographic type of ideologically moderate group over any other H11: More liberal Democrats will prefer liberal candidates to ideologically moderate candidates, but will not support one demographic type of liberal group over any other 2.6 SUMMARY AND LOOKING AHEAD This chapter began by reviewing the extant work on demographic cues as activators of stereotypes about men and women of different races, with a particular emphasis on ideology. After establishing a set of common set of assumptions that guide the theoretical and empirical 49

67 substance of the project, I develop theoretical models related to the role of demographic and policy cues in the candidate evaluation process. Given the persistence of ideological associations with race and gender, I submit that when only categorical (easy) information about partisanship and demographic status is available to primary voters, they will integrate the two in their evaluation of candidate ideology. From here, the degree to which primary voters express favorability towards different candidates should be a function of how they perceive candidates relative to their own political views. The more the two sets of views align, the higher the levels of favorability. The second part of the theory returns to the question of cue conflict in the Republican Party and how it may be resolved. I posit that since individuating (hard) information is more precise, it should overwhelm any categorical cue conflict as voters are drawn principally to the ideological substance of candidates messages. In this way, historically-underrepresented candidate groups the same ones the extant political science literature suggests are at an inherent electoral disadvantage should be able to level the electoral playing field and achieve parity with conventional white male Republican candidates. Finally, I discuss the process of evaluating counterstereotypical candidates as potentially activating positive affect for candidates once they have established ideological identities among the electorate. In the next chapters, I develop and test two survey experiments to explore these theoretical expectations. 50

68 3.0 CHAPTER THREE: METHODS AND DATA The preceding chapters established two related challenges: first, the problem the Republican Party faces in diversifying its voter base and, second, the academic puzzle of how voters will evaluate counterstereotypical Republican candidates given that the political science literature would suggest these candidates suffer from their demographic status and the psychology literature implies they may benefit from it. The second chapter explores these issues in the unique context of party primary elections which candidates must win in order to run in the fall and develops two models of voter behavior. First, the additive spatial reasoning model argues that voters will use race and gender to sort candidates within parties ideological planes and identify the best representative of their interests. Second, a variant of individuating information scale-tipping information reasons that candidates can resolve cue conflict for voters by indicating which category is more relevant in the evaluation process. Counterstereotypical candidates who can tip the scales are expected to level, and even reverse, the electoral playing field. 3.1 INVESTIGATION STRATEGY To explore these issues empirically, I develop and administer two original survey experiments administered online using Amazon s Mechanical Turk platform. This hybrid methodology 51

69 embeds a randomly assigned treatment into a survey administered to a sample of U.S. adults. This approach, in theory, allows scholars to exploit the internal validity of experimentation via random assignment with the external validity surveys offer by sampling from the overall population. Problems of artificiality, cost, and sample selection posed by experiments can be blunted (though not eliminated), as can the challenge of causal inference inherent in crosssectional survey analysis. This approach is also especially well-suited for analysis of specialized subgroups in the overall population (Mutz 2011), which any study of political primaries necessarily entails. Ultimately, as Lavine (2002, 242) put it, survey experiments that integrate representative samples with the experimental control of questions represent the most valuable tool for gaining access to the processes that underlie opinion formation. 27 Not surprisingly, scholars have increasingly relied on this approach in public opinion research (e.g. Berinsky 2007; Druckman and Leeper 2012; Sniderman and Theriault 2004; Tesler 2012; Turgeon 2009). The two survey experiments developed below differ fundamentally in the nature of the information about candidates available to primary voters. Study 1, which we may refer to as the low-information study, manipulates the race and gender of candidates while the political messages they provide to voters are both constant and ideologically neutral. This allows tests of the additive spatial reasoning model where voters are expected to use race and gender to sort their primary options. Study 2, the high-information study, manipulates race, gender, as well as policy messages. Candidates will provide primary voters with either ideologically moderate or ideologically extreme (conservative for Republicans, liberal for Democrats) messages. In both studies, analysis will focus on how voters evaluate candidates within their own party in a primary 27 But see Barabas and Jerit (2010) and Gaines, Kuklinski, and Quirk (2007) who offer important qualifications concerning the survey experiment approach. 52

70 election context. Before formalizing the two studies, however, an important discussion on the sample of voters and validation of their responses is in order. 3.2 PARTICIPANT RECRUITMENT AND VALIDATION OF RESPONSES As noted above, two survey experiments are administered through Amazon s Mechanical Turk (M-Turk) program. 28 This online platform allows scholars to can create assignments (known as Human Intelligence Tasks, or HITs) and post them on the M-Turk server. HITs are accessible by individuals, known as Turkers who have signed up to participate in the program and select from the thousands of HITs available at any given moment. Turkers, who currently number roughly 100,000, come from all walks of life and are located around the world (Pontin 2007), although I limit these studies only to American respondents. 29 The surveys themselves were designed by me using software developed by Qualtrics (a survey research company) and were subsequently administered through Qualtrics online. HITs were posted on M-Turk containing a description of the study as well as a hyperlink that took respondents to a Qualtrics webpage where the survey was posted. The description informed subjects that they would take part in a short survey [that] involves providing your opinions towards politics and society in America today. Subjects were paid $0.60 and $0.69 for the first and second surveys, respectively, for participating in one of the two HITS. 30 The M-Turk platform has become an increasingly popular method for political scientists 28 Institutional Review Board documents are available in Appendix B. 29 This is done through verification of the respondent s location based on ISP address. 30 The second study was slightly longer than the first, and thus the pay rate was marginally higher. 53

71 seeking to improve sample quality over conventional pools of college students while at the same time minimizing research costs (Mason and Suri 2012; Paolacci, Chandler, and Stern 2010). A wealth of research has been undertaken in recent years to assess the validity and utility of M- Turk as a recruitment method for social science research that helps overcome the welldocumented biases endemic to traditional college student samples (Sears 1986; Druckman and Kam 2011) or other populations representing small, unrepresentative sectors of society (Henrich, Heine, & Norenzayan, 2010). On the whole, scholars find that M-Turk samples are at least as diverse if not more representative of the population when compared to other methods for recruiting experimental/survey participants, particularly when contrasted with college student samples (Buhrmester, Kwang, and Gosling 2011). 31 Testing the external validity of M-Turk typically takes on one of two forms. First, scholars compare descriptive statistics of Turkers with respondents from traditional survey methods. Burmester, Kwang, and Gosling (2011) compare data collected from an M-Turk sample with statistics drawn from Gosling et al. s (2004) Internet-based survey, concluding that MTurk participants were more demographically diverse than standard Internet samples and significantly more diverse than typical American college samples. Berinsky, Huber and Lenz (2012) engage in a more exhaustive effort, comparing the measured characteristics of an M-Turk sample to three other types of samples: first, convenience samples used in political science research in three leading journals (American Political Science Review, American Journal of Political Science, and Journal of Politics); second, an Internet panel survey; and, third, a 31 Not all studies reach precisely the same conclusion. Berinsky, Huber and Lenz (2012), for instance, show that M- Turk improves upon convenience samples but does not quite reach the same level of representativeness as do Internet-based panels or national probability samples. They conclude that All told, [ ] the MTurk sample does not perfectly match the demographic and attitudinal characteristics of the U.S. population but does not present a wildly distorted view of the U.S. population either 54

72 probability sample of U.S. adults. They find that the M-Turk sample fares well in comparison to convenience samples, and even improves in key areas like Democratic Party identification skew (355). Similarly, compared to the Internet survey (The American National Election Panel Study (ANESP) administered by Knowledge Networks), M-Turk provides very similar results when it comes to characteristics of the measured population. 32 Critically, on key political and psychological measures, the M-Turk sample is more similar to the nationally representative samples than is the ANESP (359). 33 Turkers on the whole are somewhat more Democratic and liberal in their partisan and ideological orientations, respectively. 34 Table 2 compares the descriptive statistics of my sample with those of a nationally representative sample collected by the Pew Research Center in As the table illustrates, the M-Turk samples differ from the Pew sample in several ways. Consistent with other reviews of the Turker pool, the two studies here are comprised of individuals that are somewhat younger and more education than a nationally representative sample. Some of the differences, though, are not necessarily problematic. For instance, while white respondents are overrepresented in the M-Turk studies, my principal focus is on the Republican Party, which itself is overwhelmingly white. It is fair to conclude, then, that like other M-Turk samples those here do not constitute a perfectly representative cross-section of Americans. Still, it improves upon some other collection methods. Furthermore, since my focus is on micro-level evaluative processes (occurring within political parties) and not, say, aggregate public opinion or individual-level processes across racial minority groups, the absence of perfect representation is not as troubling. 32 There are some differences, mainly those related to major life cycle events. Turkers, for instance, are less likely to be married or own a home. 33 The M-Turk and ANESP samples were also compared to nationally representative face-to-face samples collected by the ANES and CPS. 34 This issue is not a concern for me because all respondents were blocked by party upon accepting the HIT. I set a quota for Democrats and Republicans alike and, consistent with these findings, my Democratic quota was reached more quickly than my Republican quota. 55

73 Nevertheless, it is important to recognize the bounds of any particular dataset, and the studies conducted here have their limits like any other design. 56

74 Table 2: Summary Statistics of Studies Compared to Recent National Sample Trait/Attribute Study 1 Study 2* PEW** Percent Democrats Percent Republicans Percent Female Percent White Percent Black Percent Hispanic (35.81) (38.94) (17.53) Age Range (%) (44.57) (17.62) (40.99) (17.00) (25.72) (30.45) States represented 65+ (2.00) 49 plus D.C. (North Dakota missing) 65+ (3.07) 47 plus D.C. (Alaska, South Dakota, Vermont missing) 65+ (24.78) 50 plus D.C High School or less (12.95) High School or less (11.91) High School or less (34.87) Education Level (%) Some College (23.90) Some College (37.77) Some College (20.08) College Graduate or More (51.52) College Graduate or More (50.7) College Graduate or More (40.61) Total Participants 1,050 1,076 2,002 * Note: Non-partisans and independent voters were not eligible to participate in Study 2; this affects the summary statistics for this survey experiment, most notably by inflating the percentage of Democrats and Republicans in the sample. ** Pew Research included respondents who lean Democratic or Republican as partisans. 57

75 The second approach to ensure M-Turk s validity is to replicate previous studies that do not use M-Turk and compare the results of the two approaches. Berinsky, Huber, and Lenz (2012) use M-Turk samples to replicate three experiments: first, Raskinski s (1989) classic question wording study of welfare attitudes; second, Tversky and Kahneman s (1981) framing experiment concerning an Asian Disease Problem; and, third, Kam and Simas (2010) study on risk preference and framing effects. In all three cases, the authors conclude, the experimental results found using the MTurk sample are highly similar to those found in published research (361). Given that I embed experimental treatments within the surveys administered as part of this project, it is particularly encouraging that these tests demonstrate MTurk subjects appear to respond to experimental stimuli in a manner consistent with prior research (366). Despite the research validating the M-Turk method, however, some questions about generalizability remain. 35 A critical step to ensuring sample validity is to set qualification parameters on individual HITS (Chandler, Mueller, and Paolacci 2014). Numerous steps were taken to ensure the sample for both surveys consisted of genuine, attentive partisans who were not simply advancing quickly through the survey just to achieve payment. First, only Turkers with a 90% HIT approval rating or better meaning that previous requesters were satisfied with the quality of their work and paid them at least 9 out of every 10 tasks Turkers attempted were allowed to participate. Second, only subjects with a minimum of 500 approved tasks to their credit qualified to complete the task. Third, subjects were strictly prohibited from participating more than once, and participants in one study were not eligible to participate in the other 35 For instance, in the United States, Asians are overrepresented while African American and Hispanic indivduals are underrepresented relative (Berinsky et al., 2012). Turkers also tend to be more socially anxious (Shapiro et al., 2013) and less extraverted than college-student samples (Goodman, Cryder, & Cheema, 2013; Kosara & Ziemkiewicz, 2010). 58

76 experiment. 36 Since every user has a unique identification code, I was able to identify any repeat attempts and discard them. Similarly, some potential respondents who did not initially qualify as eligible because they did not identify as partisan attempted to take the survey again, successfully surmising (or guessing) that changing their partisan affiliation may get them access to the task. I monitored the IP addresses of all participants, and in instances where multiple entries were logged from the same location in rapid succession I flagged the qualifying (second) submission and deleted it from the final data set. 37 Validation efforts were also embedded within the surveys themselves. A series of preliminary questions were used to determine a potential subject s eligibility. Only Turkers affirming they were (1) American citizens, (2) 18 years of age or older, and (3) members of a major party (Democratic or Republican) were allowed to participate. 38 Those who met these standards were allowed to complete the survey. Unseen to respondents, an electronic timer tracked how much time was spent on each question on the survey. Entries in which a participant spent less than 7 seconds on the page in Study 1 or 10 seconds on the page in Study 2 describing the political candidate (the direct treatment) were discarded because it is unlikely that the 36 Despite the inclusion of a list of Turkers who had already participated in the first study listed in the prompt for the second with a warning to check to make sure they were not on the list, a handful of Turkers attempted to take both surveys. Repeated attempted were flagged and discarded. It is also possible, however, for individuals to open multiple M-Turk accounts and repeat surveys in this fashion, although this violates Amazon s user agreement with Turkers. By monitoring ISP addresses and the timing of survey submissions, this issue can be mitigated but not completely prevented. 37 Only 20 entries from over 1,000 total Democratic and Republican subjects were discarded in this way. It should be noted also that it is technically possible for two legitimate entries (one ineligible and one eligible) to come from the same location in quick succession for instance, two college roommates make share a computer to complete MTurk tasks it is highly unlikely. There were also a few instances where multiple entries from different user IDs came from the same IP address but spread out over several days. I keep these cases since they do not appear to represent someone attempting to game the system in one sitting. It is far more plausible in these cases that separate users simply used a common computer. 38 For Study 1, I elected to allow independent and non-partisan respondents to participate, as well, to see if the additive spatial reasoning model applied to persons who do not share in-group status with the candidates in question. Since it would be prohibitively expensive to do the same in the high-information design, only partisans participated in Study 2. 59

77 participant was closely reading the material. In addition, at the end of the survey, several minutes after the treatment had been applied, respondents were asked via open-ended question to recall what they could remember about the candidate they evaluated earlier in the survey. Respondents who referenced the specific information from the treatment page the candidate s race, gender, party, policy priorities, and/or (in Study 2) ideological orientation (e.g. He was very conservative ), were retained for analysis Subject Assignment Respondents that qualified to participate in the studies were asked to read about and evaluate a candidate running in their own party s primary. For Study 1, since subjects reported their partisan affiliation in the preliminary questionnaire, they were blocked into Democrats and Republicans, respectively, and then randomly assigned to one of four candidates: a white male, a white female, an African American male, or a candidate with no demographic information. This results in two separate 3 x 3 partial factorial designs one for each party containing four treatment cells and five empty cells. A summary of this design is depicted in Table 3. A total of 512 respondents participated in the first study. Figure 6 presents the distribution of respondents across the treatment conditions On the whole, since Turkers tend to be more Democratic than the population as a whole there are typically more Democratic respondents than Republicans respondents. 60

78 Table 3: Description of Study 1 Treatment Conditions Factor 2: Demographic None (control) White African American None (control) Yes No No Factor 1: Gender Male No Yes Yes Female No Yes No All Democrats n n All Republicans 70 White male 55 Randomly assigned to: 70 White female African American Male 61 Randomly assigned to: 72 No demographic trait 54 Total Total Figure 6: Assignment of Subjects in Study 1 In the second study, partisans are again sorted into their respective party groups; this time, however, they are subsequently randomly assigned to read about one of eight candidates resulting from a 2 x 4 factorial design where the first factor is ideological tone of the message 61

79 (moderate versus extreme) and the second factor is the four demographic conditions noted above. Table 4 presents this design and Figure 7 summarizes the distribution of participants across treatment groups. Table 4: Description of Study 2 Treatment Conditions Factor 2: Demographic None (control) White Female African American Factor 1: Ideology Ideologically Moderate Ideologically Extreme Yes Yes Yes Yes Yes Yes Yes Yes 62

80 Candidate n Republicans Democrats n Candidate Moderate White male Moderate White female Moderate African American Male Moderate (No demographic trait) Conservative White male Conservative White female Conservative African American Male Randomly Randomly 70 assigned assigned 57 to: to: Moderate White male Moderate White female Moderate African American Male Moderate (No demographic trait) Liberal White male Liberal White female Liberal African American Male Conservative (No demographic trait) Liberal (No demographic trait) Total subjects Total subjects Figure 7: Assignment of Subjects in Study EXPERIMENTAL TREATMENT Manipulating Demographic Cues The manipulation of race and gender in both survey experiments was achieved by providing a picture of a white male, white female, a African American male candidate, or no picture at all. Pictures were used instead of other cues like a text box containing candidate demographic information to help ensure that respondents clearly associated a person with the message they 63

81 were reading. However, to guard against the possibility that results are artifacts of the specific pictures in the study, I utilized photographs that had been pretested and found not to vary in terms of perceptions of candidate traits. Thirty-seven undergraduates completed a short study (for course credit) in which they evaluated a series of candidates (six white males, six African American males, six white females, and six African American females) on a series of traits drawn from Todorov et al.'s (2005) study of candidate inference (these included: attractiveness, competence, likability, empathy, familiarity, willingness to vote for the candidate, typicality, and age). No ideological or partisan information was provided to the respondents. Each student was presented each picture (in a random order) and was asked to rate the individual on the traits above using Likert scales placed below the photograph. Researchers subsequently paired sets of pictures with similar overall rates and conducted difference of means tests to ensure that the two candidates were not perceived to differ significantly in terms of traits. Since racial and gender stereotypes are, as discussed in the previous chapter, related to many traits like those used in this pretest, utilizing photographs of different racial and gender groups that do not vary in terms of trait attribution will actually bias the sample against finding race and gender effects. That is, particular photos were found not to vary in terms of trait perception across white male, female, or African American candidates. As a result, I am effectively utilizing photographs which activate fairly mild or moderate sentiments when it comes to personal traits. This represents a potential drawback to this design, but since it is a more conservative approach, I opt to use it in these survey experiments. 64

82 3.3.2 Manipulating Policy Information The two studies differed fundamentally in the nature of the policy information available to primary voters. In the low-information experiment, voters were given limited information about candidates policy views. All candidates, regardless of party or race, provided the same generic policy message, tailored so as to not send an ideological signal: "I am running to be the nominee for the [Democratic/Republican] Party in the 2016 general election. I am running because I believe in America and want to do my part to help Americans thrive more than they ever have before. I pledge that I will work tirelessly to achieve important goals like providing a quality education for our children, stimulating a strong economy, and keeping our streets safe for our citizens." In the second study, respondents were randomly assigned to read one of two policy messages, either a moderate message, which was the same for both parties, or an ideologically extreme message, which for Democrats was liberal in tone and for Republicans was conservative in tone. Republican individuals, of course, evaluated either moderate or conservative candidates, while Democrats only evaluated moderate or liberal candidates. Since these studies attempt to simulate decision-making in a primary context, liberal Republicans and conservative Democrats are not included because they are not especially plausible candidates (today). Table 5 present the policy messages used in Study 2. 65

83 Table 5: Experimental Treatments Used in Study 2 Moderate Liberal Conservative As a life-long public servant, I have always believed that our job in government is to provide basic services to the people so that they may be able to pursue the American Dream. I will work to make sure that our economy is strong and citizens have jobs, that our country is safe and protected, and that our children all have access to a quality education. The best way to achieve these goals is for everyone in government to come together and pursue reasonable, balanced public policy through compromise. The best public policy is made by taking the best ideas from both parties. "As a life-long progressive, I have always believed that our job in government is to provide important services to the people to help ensure that everyone has what they need to pursue the American Dream. I will work to make sure that the government maintains a reasonable amount of control over the economy so we can ensure that people from all walks of life can find good jobs. I will work to protect American citizens by supporting diplomatic solutions to the challenges our world faces in order to prevent conflict. I will see that our children get the best education possible by supporting teachers and additional education spending in our classrooms so everyone has the resources they need to succeed. The best public policy is made by committing to core progressive ideals. As your candidate, I promise to do just that." As a life-long conservative, I have always believed that our job in government is to provide a few basic services to the public and otherwise stay out of the way so that everyone may be free to pursue the American Dream. I will work to make sure that the economy is free so that businesses can compete and grow and produce jobs. I will work to protect American citizens by ensuring that our military is the strongest in the world so it can protect our citizens. I will see that our children get the best education possible by supporting school voucher programs and reducing the negative influence of teachers unions so that school leadership is returned to the local level where it belongs. The best public policy is made by committing to core conservative ideals. As your candidate, I promise to do just that. 66

84 3.4 KEY DEPENDENT VARIABLES The hypotheses above reference primary voters evaluating candidates based on concepts like appeal and preferences. More concrete measures of affect and support are necessary for these studies. Consequently, after evaluating a political candidate, respondents in both surveys were asked to respond to three key questions. First, partisans located the candidate on a sevenpoint Likert scale running from Very liberal to Very conservative. They also indicated the degree to which they agreed with the statements that (1) they would support the candidate in an election and (2) that, if elected, the candidate would support the interests of people like themselves. For both questions, a seven-point scale was employed ( Strongly Disagree to Strongly Agree ). The ideology measure will accomplish two goals; first, it will serve as a manipulation check on the policy information. Second, it will allow me to speak to the broader literature that has focused exclusively in perceptions of candidate ideology. The latter two measures, on the other hand, are designed to look beyond perceptions of candidates to see how those impressions influence electoral behavior. Table 6 presents the three dependent variables. 67

85 Table 6: Primary Dependent Variables Dependent Variable Question posed to Respondent Scale Perception of Candidate s Ideology Which of the following do you think best describes this candidate's political ideology? Very liberal Very Conservative Voter s Willingness to Support Candidate I would support this candidate as the [Democratic/Republican] Party's nominee in the fall general election Strongly Disagree Strongly Agree Voter s Belief that Candidate Would Represent His/Her Interests If elected, this candidate would support the interests of people like me Strongly Disagree Strongly Agree 3.5 OTHER VARIABLES A variety of ancillary measures are included in both Study 1 and Study 2. A full variable index is included in Appendix A. They may be summarized here into two types of variables: 1) Demographic/personal: includes questions affirming/concerning citizenship status, state of residence, religious affiliation, marriage status, overall interest in politics, media consumption, race/ethnicity, education level, gender, income 68

86 2) Political views/status: partisanship, partisan type (weak/strong), self-identified ideology, feeling thermometers for both major parties and the Tea Party, a measure of feelings towards President Obama, voting history 3) Measures/scales of latent racism/sexism: Questions concerning latent sexism are inspired by the Sexist Attitudes Towards Women Scale (SATWS) (Benson and Vincent 1980), the Modern Sexism Scale (Swim, et al. 1995), and the Ambient Sexism Inventory (Glick and Fiske 1996). A series of questions to address latent racism are drawn from (Feldman and Huddy 2010). 4) Knowledge scale: consists of four questions which create an additive scale to serve as a proxy for political knowledge. The number of correct answers (0 to 4) will serve as an index of political sophistication. This method follows Delli-Carpini and Keeter (1996) and Gomez and Wilson (2001) among others. 3.6 SUMMARY This chapter has discussed methodology and design of the two survey experiments that constitute the empirical tests of the theory developed in the previous chapter. Efforts were made to select an appropriate method for the theoretical questions of interest, and steps were taken to validate the survey sample as much as possible. In the next two chapters, I present and discuss the results of Studies 1 and 2, respectively. 69

87 4.0 CHAPTER FOUR: LOW-INFORMATION CONTEXT The first empirical study assesses the candidate evaluation process in a low-information environment in which only categorical information (demographic status and party affiliation) about candidates is known. The theory above posits that voters are apt to use the ideological cues embedded in categorical labels (e.g. white male conservative) to sort primary candidates from one another. From here, it is the voters own ideological dispositions that are expected to influence their candidate judgments. The results, however, are mixed. They suggest that party cues may be more dominant than the theory above suggests. Indeed, it does not appear that there is much sorting of candidates within parties, even when those candidates hold demographic cues that contain ideological signals. In the following pages I document the study in detail and draw conclusions about the consequence of race and gender in low-information electoral environments. First, I present summary statistics of Republicans and Democrats and discuss the ideological distribution of partisans within those groups. This includes a brief discussion contrasting ideologues and partisans and how both are subdivided for empirical purposes. I next introduce the empirical strategy, which begins by looking at demographic cues effects on ideological perceptions and then transitions to discussing attitudinal measures. 70

88 4.1 DEFINING AND CONTRASTING IDEOLOGUES AND PARTISANS Although there is a robust correlation between a voter s ideology and the strength of her partisanship, the concepts are not analogous. The former refers to a set of ideas and beliefs that shape a person s view of the political world, and the latter concerns the nature of person s identification and association with a political party. In the analysis below my principal focus is on ideological subgroups within parties because in primaries partisanship is constant and ideological cues are expected to be the main basis by which voters sort candidates. Measures of ideology and partisanship are straightforward. Individuals were initially asked the following question: Generally speaking, do you usually think of yourself as a Republican, a Democrat, an independent or what? This facilitated initial sorting into partisan groups. Respondents were later asked to indicate their political ideology using a standard sevenpoint scale: When it comes to politics, how would you describe yourself [ ] as liberal, conservative, or neither liberal nor conservative? Respondents could select one of the following: (1) Very liberal, (2) Somewhat liberal, (3) Closer to liberal, (4) Neither liberal nor conservative, (5) Closer to conservatives, (6) Somewhat conservative, or (7) Very conservative. As Table 7 and Figure 8 show, and as we should expect, the distribution of ideologues is different across the parties, with very few strong partisans reporting strong ideological dispositions in the opposite direction In the course of cleaning the data, I dropped cases when a strong Republican reported a very liberal ideology (n=1) or a strong Democrat reported a very conservative ideology (n=4). 71

89 Table 7: Ideological and Partisan Distribution of Study 1 Participants Democrats (Overall) Weak Democrats Strong Democrats Republicans Overall Weak Republicans Strong Republicans Mean ideology (SD) 2.18 (1.09) 2.70 (1.05) 1.72 (1.04) 5.55 (1.18) 5.07 (1.12) 6.43 (0.69) Percent Very liberal (n) Percent Somewhat liberal (n) Percent Closer to liberal (n) Percent Neither liberal nor conservative (n) Percent Closer to conservative (n) Percent Somewhat conservative (n) (80) (14) (66) 0.45 (1) 0.69 (1) 0 (0) (96) (49) (47) 1.79 (4) 2.78 (4) 0 (0) (57) (49) 6.25 (8) 4.91 (11) 7.64 (11) 0 (0) 7.14 (19) (17) 1.56 (2) 5.36 (12) 7.64 (11) 1.25 (1) 3.76 (10) 5.80 (8) 1.56 (2) (71) (65) 7.50 (6) 1.50 (4) 0.72 (1) 2.34 (3) (76) (45) (31) Percent Very conservative (n) 0 (0) 0 (0) 0 (0) (49) 4.86 (7) (42) Total n

90 Figure 8: Ideological and Partisan Distribution of Study 1 Participants 4.2 CANDIDATE EVALUATIONS: REPUBLICANS Perceived Ideology: Republicans The analysis begins with a very basic question: do Republican voters perceive ideological differences across different types of candidates? The answer begins by plotting the perceived ideological location of each Republican candidates and testing to see if there are significant differences among them. Figure 9 presents a graphical representation of the average ideological score for each type of candidate, along with the average Republican (note: not to scale). As 73

91 expected the African American candidate is perceived as more liberal than the white male, but, contrary to expectations, the female candidate is viewed as slightly more conservative than the white male. The Republicans, on average, perceive themselves to be more conservative than any type of candidate. 41 These values are compared using one-way ANOVA to determine if the differences are statistically significant and the results indicate that they are not (F = 1.15, Prob > F = 0.33). Contrary to Hypothesis 2, Republicans do not perceive an ideological difference between stereotypical and counterstereotypical candidates. This finding stands in stark contrast to much of the conventional wisdom on ideological stereotypes. In this instance, the evidence suggests that voters do not impose an ideological penalty on female or African American Republican candidates, after all. They are ideologically perceived to be similar to their stereotypical counterparts. 41 Following the experiments in this project, I use the results to calculate the statistical power for each experimental condition. The results vary considerably, as some conditions generate group values very close to the sample mean, while others are much more distinct. For instance, in Study 1 sample power ranges from 5% for white Republican males to 81% for African American Republican males on the perceived candidate ideology variable. Given that there was little variation about the overall group mean for some of the test groups, particularly white males in Republican conditions, lower-than-ideal statistical power levels were occasionally achieved. For female and African Americans in the Republican Party, higher levels of power were achieved, often exceeding 80%. The distribution of overall group and treatment condition test values in this project suggest that the research will benefit from additional analysis with a larger and more representative population in future iterations. 74

92 Afr. American Candidate (4.92) White Candidate (5.12) Control Candidate (5.25)... (1) Liberal (7) Conservative Female Candidate Republicans (5.55) (5.24) Figure 9: Perceived Ideological Location of Republican Candidates To more deeply probe the issue of ideological perceptions, I assess what factors condition ideological impressions of primary candidates. To do this I estimate a regression model for each type of candidate. The key independent variable in this first set of analysis is the measure of respondents own ideological conservatism. A variety of socioeconomic and other factors that have been previously associated with candidate ideology comprise the rest of the model which is presented below and described in Table 8: Baseline OLS Model: Perceived Conservatism = β 0 + β 1 Respondent_Conservatism + β 2 Respondent_Party_Affect + β 3 Weak_Partisan + β 4 Age + β 5 White_ Respondent + β 6 Education + β 7 Female_Respondent + β 8 Income + β 9 Latent_Racism (African Americans only) + β 10 Latent_Sexism (Females only) + ε 75

93 Table 8: Description of Variables in Regression Models Dependent Variable Range Description Perceived Ideological Conservatism 1-7 Higher values correspond to higher levels of conservatism Independent Variables Respondent Conservatism 1-7 Party Affect Weak Partisan 0-1 Higher values correspond to higher levels of conservatism Higher values correspond to higher feelings of warmth for own party 1 indicates a weak partisan, 0 indicates a strong partisan Age Continuous Respondent age in year White Respondent 0-1 Education 1-8 Female Respondent 0-1 Income 0-15 Latent Racism 0-6 Latent Sexism: indicates a Caucasian respondent, 0 indicates other Higher values correspond with higher levels of education 1 indicates a female respondent, 0 indicates male Higher values correspond with higher income levels Higher values correspond with lower levels of latent racism Higher values correspond with higher levels of latent sexism 76

94 Table 9 presents the results. The effects of ideology are only significant for stereotypical white male candidates. As Republican voters grow increasingly conservative, they are increasingly likely to perceive white male primary candidates as ideologically conservative. Interestingly, the same is not true for other types of candidates. Ideological perceptions of female and African American candidates appear unaffected by the ideology of the voter evaluating them. Table 9: Determinants of Perceived Candidate Conservatism (Republicans) Control White Female African American Respondent Conservatism (0.170) (0.182)** (0.203) (0.140) Party Affect (0.008)** (0.008) (0.012)* (0.010)* Weak Partisan (0.287) (0.457) (0.312) (0.392) Age (0.011) (0.014) (0.012) (0.013) White respondent (0.64) (0.115) (0.209) (0.188) Education (0.121) (0.094) (0.143) (0.115) Female respondent (0.297) (0.306) (0.337) (0.300) Income (0.026) (0.035) (0.052) (0.045) Latent Racism (0.222) Latent Sexism (0.208) - Constant N F R RMSE

95 Figure 10 represents a graphical representation of the effects of respondent ideology on the ideological perception of the candidate they evaluate. In each panel, the X axis represents respondents ideologies ranging from (1) very liberal to (7) very conservative. The Y axis represents the predicted ideological location of each candidate using that same scale. Note that the error bars represent 90% confidence intervals, and since there are very few cases ideologically liberal Republicans, these ranges will always be larger on the leftmost part of the figure. It is clear from Figure 10 that the effects of respondent ideology are minimal in most cases; as the regression above showed, it only significantly affects perceptions of white male Republican candidates, who are perceived as increasingly ideologically conservative as respondents grow more conservative themselves. These findings may reflect some level of uncertainty about counterstereotypical candidates. When voters encounter a conventional candidate, their own ideology conditions their perception of that candidate, a phenomenon longknown to affect candidate perceptions. When they encounter unconventional candidates, though, they do not appear to impose their own ideology in the same way. Even so, this does not lead to significant overall perceptions of ideology across candidate types. Although the results are statistically insignificant, it is curious that female and African American candidates are not perceived in the same way. Female candidates have a shallow, negative slope but African American candidates, like white males, see a steady (but statistically insignificant) increase. In other words, as Republican voters grow more conservative they perceive female candidates to be more ideologically liberal (Panel C), but African American candidates are perceived to be more conservative as respondents do themselves (Panel D). To reiterate: these are statistically insignificant trends but the patterns are illustrative in that (1) respondent ideology does not matter much in evaluating the ideology of counterstereotypical 78

96 candidates perhaps owning to the cue conflict they generate and (2) the slopes in are opposite directions, implying that female and African American candidates may not be equal counterstereotypes. Panel A: Control Candidate Panel B: White Male Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Candidate Conservatism Respondent Conservatism Perceived Candidate Conservatism Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Candidate Conservatism Perceived Candidate Conservatism Respondent Conservatism Respondent Conservatism Figure 10: Predicting Candidate Conservatism As a Function of Voter Ideology (Republicans) 79

97 4.2.2 Ideological Congruence: Republicans In the last chapter I posited that respondent ideology alone was insufficient to assess candidate evaluation. If voters are seeking the most ideologically proximate candidate when making a primary election choice, then the key variable of interest should be a measure of ideological congruence, or the degree to which a voter perceives herself to be ideologically aligned with a candidate. I estimate another set of regression below but this time use ideological congruence as the dependent variable. This measure is generated by taking the absolute value of the distance between where voter i rates her own ideological position and where she rates candidate j on the same scale. The smaller the value, the less perceived ideological difference between the candidate and voter. Table 10: Average Ideological Distance Between Voter and Candidate (Republicans) Average Gap Standard Deviation Control White Female African American Before turning to the regression analysis, a preliminary examination of the data is in order. Table 10 presents the average perceived ideological gap across each demographic group. As expected, Republicans on the whole see counterstereotypical candidates as somewhat more 80

98 ideologically distant that white male candidate. There is also somewhat more variance in respondents reactions the counterstereotypical candidates than the white male. Yet when these differences are subjected to a statistical test, however, I fail to reject the null hypothesis (F = 1.00, Prob > F = 0.39). Ideological congruence is higher between Republicans and white male candidates relative to female or African American candidates, but not statistically significantly so. These early results, then, provide little support the ideological congruence hypotheses. A more thorough examination of ideological congruence begins with Table 11, which presents the regression results from this alternative specification. It shows that, as before, respondent ideology typically has little effect on the perceived ideological gap between voter and candidate. Only in the control condition was there a slightly significant effect. As Republicans grow increasingly conservative, they perceive the ideological gap between themselves and the control group Republican to grow larger. No similar results occur in any of the demographic treatment conditions. In fact, the slopes are opposite of the hypothesized direction. Figure 11 plots the effects of respondents ideology on perceived level of ideological incongruence. That is, higher values on the y-axis indicate less ideological congruence. Panel B shows, the slope for white male candidates is positive, suggesting that as Republicans grow more conservative they perceive an increasing gap between themselves and the candidate. This may be the result of more conservative voters simply moving farther and farther to the ideological right and therefore increasing the distance away from the cluster of candidates. However, the slopes for female (Panel C) and African American (Panel D) candidates are both negative, though again insignificant. The data show that more conservative Republicans do not in fact see themselves as more ideologically distant from counterstereotypical candidates with demographic cues implying 81

99 ideological liberalism. Instead, different types of ideologues see all types of candidates as relatively equidistant in terms of ideology. Table 11: Determinants of Perceived Ideological Congruity Between Voter and Candidate (Republicans) Control White Female African American Respondent Conservatism (0.294)* (0.153) (0.192) (0.124) Party Affect (0.009)** (0.006) (0.009)** (0.008) Weak Partisan (0.301) (0.357) (0.331)** (0.336) Age (0.012) (0.010) (0.011) (0.012) White respondent (0.088) (0.098) (0.265) (0.187) Education (0.101) (0.077) (0.104) (0.120) Female respondent (0.273) (0.253) (0.279) (0.270) Income (0.031) (0.031) (0.042) (0.040) Latent Racism (0.120) Latent Sexism (0.182) - Constant N F R RMSE

100 Panel A: Control Candidate Panel B: White Male Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Ideological Incongruence Perceived Ideological Incongruence Respondent Conservatism Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Ideological Incongruence Respondent Conservatism Perceived Ideological Incongruence Respondent Conservatism Figure 11: Perceived Ideological Incongruence As a Function of Voter Ideology (Republicans) Evaluating Candidates on Support and Representation Capacity: Republicans So far the theoretical expectations postulated above have not been borne out in the analysis. Republican voters tend to view Republican candidates of all demographic stripes as fairly similar to each other in terms of ideology. In this section I assess ideologically and ideological congruence as determinant of voter attitudes beyond ideological perceptions. I begin with the 83

101 issue of voters willingness to support candidates as nominees to represent the party in the general election. Table 12 presents the results. There is little effect of voters ideological perception on their willingness to support different types of candidates. The one exception, however, is again white male candidates. Here, as Republicans grow more conservative in their ideological outlook, they are increasingly less likely to support white male candidates, which is unexpected. Theoretically, more conservative Republicans should increasingly support white male candidates as their conservatism grows. While the coefficients are also negative for all the other candidates, as expected, they only achieve statistical significance with white men. Figure 12 presents the results from Table 12 in graphical form. Panel B illustrates the negative and marginally significant decline in support for white male candidates. While I cannot conclude with certainty precisely why I observe this result, it is worth noting that white male Republican candidates have the highest levels of support among ideological liberals/moderate Republican voters relative to the other candidates and therefore have further to fall since increasing levels of conservatism among Republican primary voters correspond to lower levels of support across all candidate types. While these results are inconsistent with theoretical expectations, though, they are highly consistent with those in the last section. If variance in voters attitudes towards primary candidates stems from difference in perceived ideology across candidate groups (as I originally argued), then it follows that when there are no perceived ideological differences there are subsequently few differences in other attitudes, as well. 84

102 Table 12: Support for Candidate as a Function of Ideology (Republicans) Control White Female African American Respondent Conservatism (0.271) (0.153)* (0.144) (0.114) Party Affect (0.011)* (0.006) (0.009)** (0.009)* Weak Partisan (0.343) (0.368) (0.393) (0.405) Age (0.013)* (0.011) (0.013) (0.014) White respondent (0.146) (0.188) (0.184) (0.146) Education (0.127) (0.110) (0.118) (0.114) Female respondent (0.328) (0.257) (0.328) (0.320) Income (0.032) (0.051) (0.048) (0.044) Latent Racism (0.114) Latent Sexism (0.169) - Constant N F R RMSE

103 Panel A: Control Candidate Panel B: White Male Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Support for Candidate Respondent Conservatism Support for Candidate Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Support for Candidate Support for Candidate Respondent Conservatism Respondent Conservatism Figure 12: Willingness to Support Candidate As a Function of Ideology (Republicans) The next set of analyses replace the willingness to support measure as the dependent variable with a different measure asking voters to indicate the degree to which they believe the candidate they evaluate would be a good representative of their interests if elected. Table 13 and Figure 13 present the regression analysis and predicted levels of representation capacity across the four candidate types. In no instance is there a statistically significant relationship between voters level of conservatism and their perceptions of a candidate s capacity to represent their interests. Although I theorized above that more ideologically moderate Republicans would rate 86

104 African American or female candidates as better representatives than white male candidates, this is not the case. Nor do higher levels of conservatism lead Republicans to perceive white candidates as increasingly better representatives. Consistent with all the findings so far, voters simply do not penalize counterstereotypical candidates in the way I expected they would. Table 13: Perceptions of Representation Capacity as a Function of Ideology (Republicans) Control White Female African American Respondent Conservatism (0.220) (0.149) (0.176) (0.116) Party Affect (0.009)* (0.006)** (0.009)** (0.008) Weak Partisan (0.362) (0.011) (0.341) (0.353) Age (0.015) (0.011) (0.013) (0.012) White respondent (0.078) (0.100) (0.296) (0.116) Education (0.144) (0.081) (0.120) (0.129) Female respondent (0.351)* (0.204) (0.349) (0.300)* Income (0.040) (0.037) (0.049) (0.041) Latent Racism (0.128) Latent Sexism (0.158) - Constant N F R RMSE

105 Panel A: Control Candidate Panel B: White Male Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Representation Capacity Perceived Representation Capacity Respondent Conservatism Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Representation Capacity Respondent Conservatism Perceived Representation Capacity Respondent Conservatism Figure 13: Perceptions of Representation Capacity as a Function of Ideology (Republicans) Above I argued that ideological congruence may be an even more important measure than a pure measure of ideology. To fully flesh out the congruence issue I re-estimate the last two sets of regression replacing the measure of respondents conservatism with the ideological congruence variable created earlier. As a general expectation, higher levels of congruence should lead to more favorable evaluations. In other words, when specified this way the slope for the ideological congruence variable should be negative and significant. The correlation between ideological congruence and the support and representation capacity are moderately strong at 88

106 -0.35 and -0.36, respectively. Table 14 and Figure 14 display the results. Note that unlike above, where the x-axis in each panel represents respondents level of ideological conservatism, here the x-axis presents the perceived ideological difference, or gap, between voter and candidate (e.g. 0 mean voters rate themselves as ideologically identical). Table 14: Effects of Ideological Congruence on Perceptions of Support/Represent.Capacity (Republicans) Control White Female African American Support Rep. Cap. Support Rep. Cap. Support Rep. Cap. Support Rep. Cap. Ideological Congruence Party Affect Weak Partisan Age White respondent Education Female respondent Income ** (0.152) 0.015** (0.006) (0.287) ** (0.008) (0.135) (0.128) (0.303) (0.028) ** (0.131) 0.016** (0.006) (0.307) (0.012) * (0.067) (0.140) (0.302) (0.034) ** (0.170) (0.007) (0.284) (0.011) (0.180) (0.105) (0.263) (0.046) ** (0.154) 0.015** (0.005) (0.231) (0.011) (0.072) (0.070) (0.209) (0.036) * (0.164) (0.009) (0.341) (0.013) (0.158) (0.112) (0.293) (0.040) (0.198) 0.026** (0.010) (0.294) (0.012) (0.299) (0.124) (0.323) (0.048) Latent Racism Latent Sexism (0.161) (0.139) * (0.150) (0.009) (0.353) (0.014) (0.193) (0.106) (0.299) (0.045) (0.112) (0.141) 0.014* (0.008) (0.307) * (0.012) (0.130) (0.118) * (0.286) (0.040) (0.125) - - Constant N F R RMSE

107 Consistent with expectations, in almost every case the ideological congruence coefficient is negative and significant, indicating that as the ideological gap between a voters own ideology and the perceived ideology of the candidate they evaluate decreases, voters will be more likely to (1) support that candidate and (2) view that candidate as a good representative. Interestingly, the only exceptions to this pattern occur with female and African American candidates on the issue of representation capacity. Here, a decrease in ideological congruence does not have a negative impact on how voters assess candidates as potential representatives. A closer look at Panels F and H and Figure 14 suggest, however, that this null effect may be due to the distribution of observations on the extreme end of the ideological congruence variable. Female (n=8) and African American (n=7) voters both had a handful of observations in which the perceived ideological distance between voter and candidate was greater than two. White male Republicans only hand three such cases. Other than somewhat larger error bars for African American and female candidates on the righthand side of the plot on the representation capacity variable, all the slopes in Figure 14 are roughly similar. Figure 15 presents all four candidates on one graph, and reaffirms the finding that all four candidates share a similar negative slope, but they are not statistically distinct on demographic grounds These results add additional evidence to the findings above. According to these data, but inconsistent with the theory presented in the last chapter, demographic status simply has little to no effect on candidate evaluations within the Republican Party. While some of the results were in the expected direction according to the theory (e.g. African American candidates were perceived to be the most liberal of the four; the average perceived ideological gap between voter and candidate was higher among counterstereotypical candidates), they did not by and large achieve statistical significance. The absence of even basic differences in ideological perception 90

108 across demographic groups is surprising given the consistency with which gender- and racebased ideological stereotypes are uncovered in the social science literature. Care must be taken in drawing conclusions from these data, however, particularly since they stand in stark contrast to much of the previous literature on this subject. The most charitable interpretation of these results would conclude that race and gender simply do not matter in the assessment of candidates when party information is known and constant across them. I do not draw this conclusion. There are several potential explanations for an absence of findings for instance, as noted above, the particular photographs used to indicate demographic status were intentionally selected so as to reduce the likelihood that race and gender affect perceptions. At the same time, though, if the results are valid, they contribute significantly to our understanding of how demographic cues interact with other types of information. It may be, for instance, that demographic/ideological stereotypes persist, but party membership is a more powerful cue to voters minds and therefore any negative consequences female or African American Republicans may theoretically suffer can be mediated solely through party affiliation. Additionally, it may be that the presence of other information about candidates mediates the role of race and gender in voter evaluations. Before moving on to looking at race and gender in primary elections when individuating policy information is also available to voters, however, I first assess demographic cues in Democratic voters evaluations. 91

109 Panel A: Control Candidate (Support) Panel B: Control Candidate (Rep. Capacity) Support for Candidate Predictive Margins with 90% CIs Perceived Representation Capacity Predictive Margins with 90% CIs Ideological Gap Between Voter and Candidate Ideological Gap Between Voter and Candidate Panel C: White Male Candidate (Support) Predictive Margins with 90% CIs Panel D: White Male Candidate (Rep. Capacity) Predictive Margins with 90% CIs Support for Candidate Perceived Representation Capacity Ideological Gap Between Voter and Candidate Ideological Gap Between Voter and Candidate Panel E: Female Candidate (Support) Panel F: Female Candidate (Rep. Capacity) Predictive Margins with 90% CIs Predictive Margins with 90% CIs Support for Candidate Perceived Representation Capacity Ideological Gap Between Voter and Candidate Ideological Gap Between Voter and Candidate Panel G: African American Candidate (Rep. Cap.) Panel H: African American Candidate (Rep. Cap.) Support for Candidate Predictive Margins with 90% CIs Ideological Gap Between Voter and Candidate Perceived Representation Capacity Predictive Margins with 90% CIs Ideological Gap Between Voter and Candidate Figure 14: Perceptions of Support/Rep. Capacity as a Function of Ideological Incongruence (Republicans) 92

110 Panel A: Support for Candidate Predicted Level of Support Ideological Gap Between Voter and Candidate Control Female White_Male African_American Panel B: Perceived Representation Capacity Predicted Level of Support Ideological Gap Between Voter and Candidate Control Female White_Male African_American Figure 15: Perceptions of Support/Rep. Capacity as a Function of Ideological Congruence (Republicans) 93

111 4.3 CANDIDATE EVALUATIONS: DEMOCRATS Perceived Ideology: Democrats The absence of results for Republican voters makes it all the more important to consider Democrats, as well. If no significant demographic effects occur here, I have more reason to revisit the experimental design, or, potentially, I have a stronger basis for arguing that it is possible demographic considerations do not carry over into party politics. The analysis in this section mirrors that above. First, I plot the perceived ideological location of each type of Democrat, shown in Figure 16 (Not to scale). As expected, the African American candidate was perceived to be the most liberal, roughly 0.35 points more liberal than other candidates, who are all clustered around the 3.30/7. However, as with Republicans, this difference is not statistically significant. An examination using one-way ANOVA comparing means across all four groups show demographic status does not affect ideological perceptions among Democrats, either (F = 1.56, Prob > F = 0.200). 94

112 Afr. American Candidate (2.94) White Candidate (3.30)... (1) Liberal (7) Conservative Democrats (2.23) Female Candidate (3.29) Control Candidate (3.32) Figure 16: Perceived Ideological Location of Democratic Candidates (All Democrats) Table 15 presents the results from the regression models estimating the effects of voter ideology on perceived candidate ideology. Consistent with literature showing the individuals tend to impose their own ideological views on candidates they evaluate, there is a positive slope for all four candidates on the respondent conservatism coefficient. That is, across all four candidates, although only marginally significant for two of them, white male and female Democrats, as voters grow increasingly conservative in their own ideology the more conservative they perceive candidates to be. Figure 17 presents these results in graphical form. As with Republicans, Democratic voters tend to rate candidates ideologically on the basis of their own political views and not on the demographic group to which the candidates belong. This result is somewhat surprising given the persistence of racial and gender effects in a variety of contexts, but it is somewhat less surprising than the similar null findings among Republicans. 95

113 Since there are no counterstereotypical candidates in the Democratic field, and since all cues are consistent with a more liberal ideology, it follows that there is a strong association between voter ideology and his perceptions of candidate ideology. Table 15: Determinants of Perceived Candidate Conservatism (Democrats) Control White Female African American Respondent Conservatism (0.116) (0.145)* (0.137)* (0.179) Party Affect (0.005)** (0.012) (0.007) (0.010)** Weak Partisan (0.322) (0.392) (0.426) (0.432) Age (0.010) (0.016) (0.013)* (0.013) White respondent (0.082) (0.163) (0.130) (0.081) Education (0.080) (0.114)* (0.144) (0.110) Female respondent (0.235) (0.286) (0.350) (0.317) Income (0.036) (0.045) (0.046) (0.036) Latent Racism (0.178)* Latent Sexism Constant N F R RMSE

114 Panel A: Control Candidate Panel B: White Male Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Candidate Conservatism Respondent Conservatism Perceived Candidate Conservatism Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Candidate Conservatism Respondent Conservatism Perceived Candidate Conservatism Respondent Conservatism Figure 17: Predicting Candidate Conservatism As a Function of Voter Ideology (Democrats) Ideological Congruence: Democrats As with Republicans, ideology should be assessed not only in raw terms but in terms of congruence between voter and candidate. I again calculate an ideological congruence variable, generated by taking the absolute values of the difference between where a voter locates herself 97

115 on a seven-point ideological scale and where she places the candidate she evaluated. Table 16 displays the average perceived ideological gap between Democrats and each type of candidate. Consistent with expectations, the smallest differences (indicating highest levels of ideological congruity) occur among female and African American candidates. Yet the difference in differences between white male candidates/voters and African American and female candidates/voters is relatively small in magnitude. Not surprisingly, the effect is statistically insignificant when subjected to ANOVA tests (F = 0.32, Prob > F = 0.81). The initial analysis of ideological congruity, then, provides no evidence to support the claim that Democrats see themselves as particularly ideologically proximate to more liberal looking candidates. Table 16: Average Ideological Distance Between Voter and Candidate (Democrats) Average Gap Standard Deviation Control White Female African American Still, Table 16 cannot uncover any effects of differences in ideology among Democratic voters when it comes to evaluating candidates. For that I turn to an additional set of regression analysis in which the dependent variable is ideological incongruence rather than a simple measure of ideology. Theoretically, I expect that the slope should be negative and significant for white male candidates, indicating that as Democrats grow more conservative, they will increasingly perceive themselves as ideologically similar to white male candidate (i.e. low 98

116 incongruence). The slopes for female and African American candidates should be positive, since highly liberal respondents should see a smaller ideological gap between themselves and female or African American candidates than they would when evaluating white male candidates. Table 17: Determinants of Perceived Ideological Congruity Between Voter and Candidate (Democrats) Control White Female African American Respondent Conservatism (0.141)** (0.141)* (0.155)** (0.192) Party Affect (0.006)** (0.012) (0.007) (0.011) Weak Partisan (0.253)** (0.270) (0429) (0.387) Age (0.012) (0.017) (0.014) (0.014) White respondent (0.087) (0.132) (0.119)** (0.075) Education (0.080) (0.121) (0.150) (0.118) Female respondent (0.248) (0.277)* (0.390) (0.303) Income (0.035) (0.040) (0.039) (0.047) Latent Racism (0.215) Latent Sexism (0.205) - Constant N F R RMSE

117 The data, however, do not reveal this to be the case. Although an increase in conservatism among Democrats corresponds with a significant decrease in the perceived ideological gap with white male candidate in Panel B of Figure 18, in support of Hypothesis 3d, this is not an isolated affect. The evidence reveals that this negative slope emerges across all candidates. Only in the African American candidate condition does this affect not reach statistical significance. The results are consistent with the pattern of demographic status having little bearing on ideological inference about candidates. Democratic voters, not surprisingly, systematically perceive their ideological incongruence with all candidates to increase as the voters themselves grow more conservative. Somewhat more surprising, though, is that demographic status does little to affect this process. It simply supports the contention that regardless of what a candidate looks like, more conservative Democrats will view that candidate as more ideologically incongruent than they will more moderate party candidates. 100

118 Panel A: Control Candidate Panel B: White Male Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Ideological Inongruence Perceived Ideological Inongruence Respondent Conservatism Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Ideological Inongruence Respondent Conservatism Perceived Ideological Inongruence Respondent Conservatism Figure 18: Perceived Ideological Congruence As a Function of Voter Ideology (Democrats) Evaluating Candidates on Support and Representation Capacity: Democrats If demographic status does not affect ideological impressions, I noted above, it follows that subsequent attitudes may not be affected. This was the case among Republican voters. Here I look at race and gender in Democratic contests to see support and representation capacity are similarly unaffected demographic information given what has been observed thus far in the 101

119 results. Table 18 displays the results of models predicting support for different types of Democratic candidates. The original theory would have more ideologically conservative voters expressing more support for white male candidates than female or African American candidates. This does not appear to be the case. The coefficient for respondent ideology is highly insignificant in the white male model yet in the female condition it is significant and positive, which was unexpected. The results suggest that Democrats are more inclined to support female party candidates as the voters themselves grow more conservative. Moving from the most liberal to the most conservative Democrats, predicted levels of support increase almost two points on a seven-point scale (4.77 to 6.55). Moreover, size of the respondent conservatism variable for the female candidate is dramatically larger than either other demographic group. This raises a cautionary flag, for there is little theoretical reason for women to be perceived so differently from other demographic groups. Even if this effect is artifactual, however, the larger point remains: Overall, Democratic voters are not more inclined to support one demographic group over another on the basis of voter ideology. A final point worth making on the support issue is to reiterate a comment made earlier: in this instance female and African American candidates have different slopes positive for women, negative for African Americans (Panels C and D in Figure 19). While the effect of ideology is insignificant in the latter case, the results again suggest that I must be careful in theoretically treating women and African Americans as analogous entities in this context. While their results are usually similar, and while demographic status usually does not have a bearing on attitudes towards candidates, there are clearly some instances in which voters may react differently to different demographic cues. 102

120 Table 18: Support for Candidate as a Function of Ideology (Democrats) Control White Female African American Respondent Conservatism (0.095)* (0.120) (0.149)** (0.157) Party Affect (0.005)** (0.007)** (0.007)* (0.010) Weak Partisan (0.262) (0.251) (0.461) (0.329) Age (0.009) (0.012) (0.013) (0.009) White respondent (0.104) (0.138) (0.126) (0.084) Education (0.070) (0.108) (0.135) (0.100) Female respondent (0.220)* (241) (0.314) (0.248) Income (0.031) (0.035) (0.043) (0.037) Latent Racism (0.182) Latent Sexism (0.194) - Constant N F R RMSE

121 Panel A: Control Candidate Panel B: White Male Candidate Support for Candidate Predictive Margins with 90% CIs Support for Candidate Predictive Margins with 90% CIs Respondent Conservatism Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Support for Candidate Predictive Margins with 90% CIs Respondent Conservatism Support for Candidate Predictive Margins with 90% CIs Respondent Conservatism Figure 19: Willingness to Support Candidate As a Function of Ideology (Democrats) I turn next to the issue of representation capacity. Not surprisingly, there are few results to speak of across candidate groups. Interestingly, the only time respondent ideology has a significant effect is in the female candidate treatment condition, and again the effect is a positive and moderately significant slope. Table 19 presents the regression results and Figure 20 presents the plots of predicted level of representation capacity. As with the support variable, the more conservative a Democratic respondent, the higher the perceived representation capacity but 104

122 only for female candidates! Respondent ideology does not condition the attitudes towards representation for any other type of candidate. In fact, not only are the results statistically insignificant for both white and African American men, but their slopes slightly negative. Coupled with the findings concerning support above, these results suggest there may be something about the female candidate that is appealing to more conservative Democrats. The underlying cause of this pattern may be related to this particular treatment, or there may be another explanation not predicted here. Additional analysis in future experiments will help tease out why I observe these curious results. 105

123 Table 19: Perceptions of Representation Capacity as a Function of Ideology (Democrats) Control White Female African American Respondent Conservatism (0.089) (0.133) (0.139)* (0.173) Party Affect (0.005)** (0.011)** (0.006) (0.012) Weak Partisan (0.250) (0.318) (0.412) (0.351) Age (0.009)* (0.014) (0.012) (0.010) White respondent (0.066) (0.139) (0.157) (0.091) Education (0.057) (0.141) (0.119) (0.092) Female respondent (0.214) (0.250) (0.293) (0.269) Income (0.028) (0.052) (0.058) (0.037) Latent Racism (0.210)* Latent Sexism (0.154) - Constant N F R RMSE

124 Panel A: Control Candidate Panel B: White Male Candidate Perceived Representation Capacity Predictive Margins with 90% CIs Respondent Conservatism Perceived Representation Capacity Predictive Margins with 90% CIs Respondent Conservatism Panel C: Female Candidate Panel D: African American Candidate Predictive Margins with 90% CIs Predictive Margins with 90% CIs Perceived Representation Capacity Perceived Representation Capacity Respondent Conservatism Respondent Conservatism Figure 20: Perceptions of Representation Capacity as a Function of Ideology (Democrats) A final set of models assess the role of ideological incongruity on attitudes towards Democratic primary candidates. As with Republicans these results are combined into a single table Table 20. Recall that when respondent ideology is replaced with a measure of ideological congruence, the coefficient should generally be negative and significant, indicating that as the perceived ideological gap between voter and candidate increases, support for candidates and perceptions of representation capacity should go down. This is precisely what we 107

125 observe. With one exception, the ideological incongruence measure is indeed negative and significant across all candidates. Figure 21 shows a series of predicted levels of support and representation capacity across demographic groups and they are roughly similar throughout. Moreover, when any combination of candidates are plotted on the same figure the slopes are parallel but statistical significance across demographic types never emerges. Figure 22 overlays the prediction slopes for all four candidates. The pattern is similar for all four. As with the rest of this chapter, the results here suggest that it is the voters not the candidates who have the greatest impact on attitudes towards candidates. Since there are no counterstereotypical candidates in the Democratic Party, it is somewhat more plausible that few results emerge. However, they still roundly refute the theory that demographic cues lead to ideological subtyping within parties. 108

126 Table 20: Effects of Ideological Congruence on Perceptions of Support/Represent.Capacity (Democrats) Control White Female African American Support Rep. Cap. Support Rep. Cap. Support Rep. Cap. Support Rep. Cap. Ideological Congruence Party Affect Weak Partisan Age White respondent Education Female respondent Income ** (0.098) 0.008** (0.004) (0.249) (0.007) (0.085) (0.059) (0.202) (0.025) ** (0.086) 0.014** (0.004) (0.234) 0.017** (0.008) (0.066) (0.056) (0.185) (0.024) * (0.109) 0.032** (0.007) (0.264) (0.012) (0.139) (0.123) (0.245) (0.034) ** (0.124) 0.019** (0.008) (0.312) (0.013) (0.140) (0.129) (0.256) (0.052) ** (0.103) 0.013** (0.005) (0.278) (0.010) (0.104) (0.100) (0.254) (0.031) ** (0.115) (0.005) (0.289)) (0.010) * (0.138) (0.094) (0.270) (0049) Latent Racism Latent Sexism (0.132) (0.108) * (0.159) (0.009) (0.300) (0.007) (0.083) (0.069) ** (0.220) (0.039) (0.135) (0.120) (0.011) (0.366) (0.010) (0.105) (0.084) (0.239) (0.041) ** (0.176) - - Constant N F R RMSE

127 Panel A: Control Candidate (Support) Panel B: Control Candidate (Rep. Capacity) Predictive Margins with 90% CIs Predictive Margins with 90% CIs Support for Candidate Perceived Representation Capacity Ideological Gap Between Voter and Candidate Ideological Gap Between Voter and Candidate Panel C: White Male Candidate (Support) Panel D: White Male Candidate (Rep. Capacity) Predictive Margins with 90% CIs Predictive Margins with 90% CIs Support for Candidate Perceived Representation Capacity Ideological Gap Between Voter and Candidate Ideological Gap Between Voter and Candidate Panel E: Female Candidate (Support) Panel F: Female Candidate (Rep. Capacity) Predictive Margins with 90% CIs Predictive Margins with 90% CIs Support for Candidate Perceived Representation Capacity Ideological Gap Between Voter and Candidate Ideological Gap Between Voter and Candidate Panel G: African American Candidate (Rep. Cap.) Panel H: African American Candidate (Rep. Cap.) Support for Candidate Predictive Margins with 90% CIs Ideological Gap Between Voter and Candidate Perceived Representation Capacity Predictive Margins with 90% CIs Ideological Gap Between Voter and Candidate Figure 21: Perceptions of Support/Rep. Capacity as a Function of Ideological Incongruence (Democrats) 110

128 Panel A: Support for Candidate Predicted Level of Support Ideological Gap Between Voter and Candidate Control Female White_Male African_American Panel B: Perceived Representation Capacity Predicted Level of Support Ideological Gap Between Voter and Candidate Control Female White_Male African_American Figure 22: Perceptions of Support/Rep. Capacity as a Function of Ideological Incongruence (Democrats) 111

129 4.4 CONCLUSIONS: A CASE OF MINIMAL DEMOGRAPHIC EFFECTS Since Democrats of different demographic stripes arguably represent less cue conflict than similar candidates do for Republicans, it may not be particularly surprising to observe little variation in attitudes and preferences within the Democratic Party. Yet to find almost no demographic effects in the Republican Party, as well, is a curious observation. Across both parties, a candidate s demographic status did not affect individuals perceptions either of candidate ideology or ideological congruence between voter and politician. Nor did demographic status impact voters willingness to support candidates or their perceptions of candidates ability to represent them. Critically, contrary to expectations, African American and female Republicans were not perceived to be more ideologically aligned candidate options than white male candidates among the more moderate and liberal members of the Republican Party. Nor were those same candidates rated more favorably by increasingly liberal members of the Democratic Party. The results overall suggest that ideological congruence matters far more that demographic status as an informational cue. Across all candidates, as voters perceived higher levels of ideological incongruence between themselves and the candidates they evaluated, their impressions of those candidates grew, predictably, less favorable. Yet the slopes for each demographic type of candidate in both parties statistically overlapped quite a bit, reinforcing the conclusion that demographic effects are simply not emerging as expected. A natural question to emerge from this analysis is why are there no demographic effects in this experiment despites decades of research concluding that demographic stereotypes are routinely applied when evaluating politicians (or others)? As Sanbonmatsu and Dolan (2009, 485) conclude, the presence of a party cue does not preclude a role for candidate gender. Yet 112

130 the results here along with those in some other studies (e.g. Hayes 2011; Huddy and Capelos 2002) fail to unearth gender (or race) effects. There are several possible explanations. First, as noted above, the treatment may not have been sufficiently stimulating. By utilizing bland photographs to depict different demographic groups, it may simply be that the pictures did not convey a strong enough sense of race and gender. This is one downside to a conservative treatment strategy it increases the potential for a Type II error. There are other possibilities, though. An alternative interpretation would be that partisan information has an equalizing effect across candidate groups. It may be that the presence of partisan cues overwhelms demographic stereotyping that is in fact happening. I cannot definitely draw this conclusion using these data, however, since partisanship never varied in any of the treatment conditions. Ultimately, in the debate over how demographic and partisan cues interact to affect voter behavior, the results from this chapter contribute additional evidence to the position that demographic information is not particularly effective in inducing shifts in evaluations of political candidates. Yet I again caution that these results do not settle the issue. Indeed, the absence of effects in this experiment make it all the more necessary to continue to probe this issue empirically. In the next chapter I take steps to begin to do this. 113

131 5.0 CHAPTER 5: IDEOLOGICAL PERCEPTIONS IN A HIGH-INFORMATION CONTEXT The second survey experiment expands on the first one by adding another element policy information to the evaluative process. The results illustrate that partisan voters clearly distinguish different ideological types of candidates within their own parties despite shared party status and these distinctions have significant consequence for candidate inference. Below I focus on the interaction between demographic cues and ideological policy information within political parties, assessing their effects on ideological perceptions of different types of candidates. Recall that counterstereotypical status is theorized to make little difference in a highinformation environment, and this in fact turns out to be the case: the results show that variations in ideological perceptions are largely driven by (1) the tone of candidates messages and (2) the nature of voters ideology, just as the results in the last chapter. 5.1 MANIPULATION CHECK The analysis begins with a manipulation check of the information treatment conditions. This is done by comparing the control group candidates (with no demographic information) across their ideological subgroups. Overall, the manipulations were successful. Republicans identified the conservative party candidate as significantly more conservative than the moderate Republican (t 114

132 = 10.39, p = 0.000), and Democrats perceived the liberal candidate to be significantly more liberal than the moderate Democrat (t = 2.72, p = 0.007). Despite a common party label, then, partisans overall recognized some party candidates as ideologically distinct from others within their party based on policy information. Moreover, these results hold regardless of demographic cues; African American, white, male, or female, conservative (liberal) policy information leads respondents to view candidates as considerably more conservative (liberal) than moderate policy information. These results represent a positive start, for they affirm that voters recognize and distinguish ideological substance in candidate messages over and beyond party identification. I can now turn to more explicit tests of the interaction between policy and demographic information. 5.2 THE EFFECTS OF DEMOGRAPHIC CUES AND POLICY INFORMATION ON IDEOLOGICAL PERCEPTIONS: REPUBLICANS In the theory above I argued the policy information draws individuals away from cue conflict, and therefore when this information is available voters will not perceive significant ideological differences across demographic groups when candidates all share the same policy information (Hypothesis 7). To assess this question I begin by presenting results from two-way ANOVA tests in which I interact the policy factor (moderate, conservative) with the demographic factor (control, white, female, African American). Since the number of interactions created by doing this is large (28), I focus for the moment only demographic comparisons within ideological groups (e.g. moderate candidates vs. moderate candidates). Theoretically, a difference in perceived ideology should not differ. 115

133 Table 21: Perceptions of Candidate Ideology Across Ideological Subgroups: Republicans Comparison Contrast Standard error T White Male (moderate) vs. Female (moderate) White Male (moderate) vs. African American (moderate) Female (moderate) vs. African American (moderate) White Male (conservative) vs. Female (conservative) White Male (conservative) vs. African American (conservative) Female (conservative) vs. African American (conservative) The results in Table 21 are consistent with expectations. In no instance is there a significant difference between any pair of candidates sharing similar ideological views. While there are main effects for ideology (F = , Prob > F = 0.000), there are no main effects for the demographic factor (F = 1.33, Prob > F = 0.234) or the interaction of the two (F = 1.13, Prob. > F = 0.336). Yet while the evidence supports Hypothesis 7, it is important to bear in mind the demographic cues did not affect ideological perceptions in the low-information environment, either. It is possible that these results, like those above, may be artifacts of, for instance, weak treatments. Nevertheless, the evidence here is at least suggestive that any theoretical differences in how difference gender and racial groups are ideologically evaluated may dissolve when specific ideological information is available to voters. One implication of these results is that counterstereotypical candidates may not only achieve ideological parity with conventional ones through their policy messages, but they may be able to assert themselves as significantly more conservative if they run against moderate white 116

134 males. To assess this possibility I present another set of comparisons from the two-way ANOVA, this time comparing different demographic and ideological groups. Table 22 presents the results from this set of comparisons. Unlike before, these comparisons are all statistically significant, indicating that not only are there no effect of demographic status within ideological subgroups but counterstereotypical candidates can ever appear more conservative than white male candidates given the right combination of policy agendas. Table 22: Perceptions of Candidate Ideology Across Ideological and Demographic Subgroups: Republicans Comparison Contrast Standard error T White Male (moderate) vs. Female (conservative) White Male (moderate) vs. African American (conservative) Female (moderate) vs. African American (conservative) White Male (conservative) vs. Female (moderate) White Male (conservative) vs. African American (moderate) Female (conservative) vs. African American (moderate)

135 5.2.1 Assessing Race, Gender, and Policy Information on Perceived Candidate Ideology The two-way ANOVA provides a reasonable start to the analysis, but, as in the previous chapter, they do not explicitly account for the ideology of Republican participants. While correlation between ideology and partisanship is modest (0.50) the two variables must be considered distinctly in the theoretical model. Thus I again employ regression analysis that will allow me to account for variation in evaluations as a function of both of these considerations. I begin with an overall assessment of the role of voter characteristics in the ideological perception process by analyzing the control group candidates only. One model is estimated for moderate candidates and a second is estimated for conservative candidates. The results are presented in Table 23 and Figure 23, respectively. Like the ANOVA results above, policy information is the driving force behind differences in ideological perceptions across candidates. Figure 23 affirms that individuals tend to impose their own ideology on perceptions of candidates. As Republicans grow more conservative, they tend to view conservative candidates increasingly conservative and moderate candidates as increasingly liberal. The initial regression results, in short, support the pattern that has emerged thus far: policy information, in conjunction with voters own ideological disposition, conditions ideological perceptions of candidates. 118

136 Table 23: Perceptions of Candidate Ideology in Control Condition (Republicans) Control Moderate Conservative Respondent Conservatism (0.193) (0.123) Party Affect (0.009)** (0.005) Weak Partisan (0.378) (0.229) Age (0.014) (0.008) White respondent (0.630) (0.272) Female respondent (0.346) (0.234) Education (0.111) (0.079) Income (0.047) (0.026) Constant N F R RMSE

137 Perceived Candidate Conservatism Respondent's Conservatism Conservative Moderate Figure 23: Perceptions of Control Group Candidate Ideology by Policy Message (Republicans) Table 24: Perceptions of Ideological Congruence in Control Condition (Republicans) Control Moderate Conservative Respondent Conservatism (0.163) (0.131)** Party Affect (0.008)** (0.005) Weak Partisan (0.327) (0.234) Age (0.112) (0.007) White respondent (0.623) (0.341) Education (0.106) (0.077) Female respondent (0.301) (0.196) Income (0.048) (0.027) Constant N F R RMSE

138 Ideological Incongruence Respondent's Conservatism Conservative Moderate Figure 24: Perceptions of Ideological Congruence in Control Condition by Policy Message (Republicans) Before moving to the analysis of race and gender I also assess the role of policy information in terms of ideological congruence within the control condition. In this instance, the slope for conservative candidates should be negative and significant while the slop for moderate candidates should be positive and significant. That is, as Republicans grow more conservative, they should perceive a smaller ideological difference between themselves and conservative candidates but perceive a larger ideological difference between themselves and moderate candidates. Table 24 and Figure 24 present the results, which are exactly as anticipated. Ideological congruence with conservative (moderate) candidates increases (decreases) as Republican voters grow more conservative. The results are consistent with a model of decisionmaking in which voters use ideological information to assess the degree to which they ideologically align with primary candidates. In the next sections, I extend this analysis to integrate demographic information. 121

139 Policy Information, Gender, and Perceptions of Candidate Ideology In this section of the analysis I compare counterstereotypical candidates separately. In the first set of regressions, African American candidates are dropped from the analysis and male and female Republican candidates are assessed relative to the baseline control group. In the next section, female candidates are omitted and white and African American men are compared. This division of candidates is useful for two reasons: first, it more accurately reflects Republican primaries in that to the extent that counterstereotypical candidates are present there is often just one (e.g. a female or an African American). Second, this step allows me to compare two groups more explicitly that are theoretically expected to generate similar outcomes. Recall, for instance, that in the last chapter there were instances of African American and female candidates generating different empirical outcomes. In Table 25 I present a pair of regressions, one for the moderate and one for the conservative. The control group candidates are included to facilitate separate estimates of male and female candidates. Following the estimation of the model predicted ideology values are generated for each of the four candidate types (Moderate Male, Conservative Male, Moderate Female, Conservative Female) and Figure 25 presents a series of comparison across them. The results show that policy information continues to be the principal determinant of candidate ideology. The results here mirror those in the two-way ANOVA above in every pairwise comparison of moderates vs. conservatives regardless of the gender of the two candidates contrasted the conservative candidate is perceived to be significantly more conservative than the moderate candidate (Panels A D in Figure 25). Similarly, when men and women sharing the same ideology are compared, perceived ideology is unaffected by gender. Here, however, 122

140 the effect is shown to hold up while controlling for respondent ideology. As above, slopes for all candidates are positive, consistent with the idea that voters project their own ideology onto candidates they evaluate. The results here are remarkably similar to those above when only control candidates were analyzed, and facilitate the interpretation that policy information, not gender, is the basis for ideologically discriminating between male and female Republican primary candidates. As theorized, gender effects disappear in the high-information environment. 123

141 Table 25: Perception of Male and Female Ideology by Policy Message Type (Republicans) Moderate Conservative Respondent Conservatism (0.113)* (0.091) Party Affect (0.005)** (0.003) Weak Partisan (0.229) (0.189) Age (0.007) (0.006)** White respondent (0.357)* (0.373) Shared Gender (0.236) (0.203) Education (0.073) (0.067) Income (0.026) (0.020) Latent Sexism (0.114) (0.132) Female Candidate (0.250) (0.161) Male Candidate (0.245) (0.241) Constant N F R RMSE

142 Panel A: Conservative Female vs. Moderate Female Panel B: Conservative Male vs. Moderate Male Perceived Ideology of Candidate Perceived Ideology of Candidate Respondent's Conservatism Conservative_Female Moderate_Female Respondent's Conservatism Conservative_Male Moderate_Male Panel C: Conservative Female vs. Moderate Male Panel D: Conservative Male vs. Moderate Female Perceived Ideology of Candidate Perceived Ideology of Candidate Respondent's Conservatism Conservative_Female Moderate_Male Respondent's Conservatism Moderate_Female Conservative_Male Panel E: Conservative Female vs. Conservative Male Panel F: Moderate Female vs. Moderate Male Perceived Ideology of Candidate Perceived Ideology of Candidate Respondent's Conservatism Respondent's Conservatism Conservative_Female Conservative_Male Moderate_Female Moderate_Male Panel G: All Candidates Perceived Ideology of Candidate Respondent's Conservatism Conservative_Female Conservative_Male Moderate_Female Moderate_Male Figure 25: Perception of Male and Female Ideology by Policy Message Type (Republicans) 125

143 Policy Information, Race, and Perceptions of Candidate Ideology The same analysis is repeated with female candidates dropped from the sample and white male and African American male candidates contrasted relative to control group candidates. The model is largely the same, although this time the demographic indicator represents shared race between respondent and candidate rather than shared gender above. The results, however, do not change. Policy information and respondent ideology condition ideological interpretations. Figure 26 below displays plots that are nearly identify to those in the last figure: when different ideological groups are compared, voters perceive moderates and conservatives differently, regardless of race (Panels A-D). Simultaneously, when ideology is constant across candidate group (Panels E-F), there are no perceived differences on the basis of race. Consistent with expectations, then, these results show that policy information continues to be the central consideration, across both gender and race. 126

144 Table 26: Perception of White and African American Ideology by Policy Message Type (Republicans) Moderate Conservative Respondent Conservatism (0.109) (0.090)* Party Affect (0.006)** (0.004) Weak Partisan (0.273) (0.209) Age (0.008) (0.005) Shared Race (0.617) (0.738) Female respondent (0.210) (0.158) Education (0.067)* (0.062) Income (0.025)** (0.021) Latent Racism (0.105) (0.069) White Candidate (0.628) (0.747) African American Candidate (0.231) (0.152)** Constant N F R RMSE

145 Panel A: Conservative White vs. Moderate White Panel B: Conservative AfrAm vs. Moderate AfrAm Perceived Ideology of Candidate Perceived Ideology of Candidate Respondent's Conservatism Conservative_White Moderate_White Respondent's Conservatism Conservative_AfrAm Moderate_AfrAm Panel C: Conservative White vs. Moderate AfrAm Panel D: Moderate White vs. Conservative AfrAm Perceived Ideology of Candidate Respondent's Conservatism Perceived Ideology of Candidate Respondent's Conservatism Moderate_White Conservative_AfrAm Conservative_White Moderate_AfrAm Panel E: Conservative White vs. Conservative AfrAm Panel F: Moderate White vs. Moderate AfrAm Perceived Ideology of Candidate Perceived Ideology of Candidate Respondent's Conservatism Conservative_White Conservative_AfrAm Respondent's Conservatism Moderate_White Moderate_AfrAm Panel G: All Candidates Perceived Ideology of Candidate Respondent's Conservatism Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Figure 26: Perception of White and African American Ideology by Policy Message Type (Republicans) 128

146 5.2.2 Assessing Race, Gender, and Policy Information on Perceived Ideological Congruence As with all the previous analysis the results here are subjected to tests of ideological congruity and not simply measures of ideology. If information remains as central to the evaluation process as it has thus far, we should expect to that ideological congruence should vary as a function of voter ideology, just as it did above in the control group-only setting. The difference here is as Republicans grow more conservative, they should perceive a smaller ideological difference between themselves and male or female (African American) conservative candidates but perceive a larger ideological difference between themselves and male or female (African American) moderate candidates Gender, Policy Information, and Ideological Congruence The results stand as expected. When candidates of different genders offer similar policy messages, voters use that information to determine ideological congruity between themselves and their primary option. Table 27 present results for male and female candidates in both a moderate condition and conservative condition. It follows that among that in the population of moderate candidates, voters perceive an increasingly large ideological gap as their own ideology grows more conservative. Similarly, among conservative candidates, the respondent conservatism coefficient is negative and significant. Critically, however, as Figure 27 demonstrates, this pattern holds both within and between candidate gender types. Panels A and B compare moderates and conservatives of the same gender while Panels C and D compare male and female candidates across ideology. In all four cases the slope for moderate candidates is 129

147 positive (as expected), and the slope for conservative candidates is negative (as expected). In short, policy information remains the principal determinant of ideological perceptions, both in terms of raw ideology levels as well as ideological congruence between voter and candidate. Table 27: Perceptions of Ideological Incongruence With Male and Female Candidates by Policy Message Type (Republicans) Moderate Conservative Respondent Conservatism (0.090)** (0.092)** Party Affect (0.005)** (0.003)* Weak Partisan (0.207)* (0.172) Age (0.007) (0.005) White respondent (0.344) (0.288) Shared Gender (0.220) (0.164)** Education (0.067) (0.056) Income (0.026) (0.019) Latent Sexism (0.102) (0.119) Female Candidate (0.237) (0.147) Male Candidate (0.237) (0.692) Constant N F R RMSE

148 Panel A: Conservative Female vs. Moderate Female Panel B: Conservative Male vs. Moderate Male Perceived Ideological Incongruence with Candidate Respondent's Conservatism Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_Female Moderate_Female Conservative_Male Moderate_Male Panel C: Conservative Female vs. Moderate Male Panel D: Conservative Male vs. Moderate Female Perceived Ideological Incongruence with Candidate Respondent's Conservatism Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_Female Moderate_Male Conservative_Male Moderate_Female Panel E: Conservative Female vs. Conservative Male Panel F: Moderate Female vs. Moderate Male Perceived Ideological Incongruence with Candidate Respondent's Conservatism Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_Male Conservative_Female Moderate_Female Moderate_Male Panel G: All Candidates Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_Female Moderate_Female Conservative_Male Moderate_Male Figure 27: Perceptions of Ideological Incongruence by Gender and Policy Message Type (Republicans) 131

149 Assessing Race and Policy Information on Perceived Ideological Congruence The same analysis is conducted replacing the gender-model with a race-based model. The results, presented in Table 28, are nearly identical to those in the last section with one curious exception. While the findings do not change overall, here we see some of the first evidence of demographic information affecting ideological perceptions. The effect occurs with moderate African American Republican candidates. Note that in Panel B (and C) of Figure 28 there is a negative and significant slope for conservative African American (white) Republicans, indicating that as Republicans grow more conservative they view themselves as increasingly aligned with these candidates. Yet among moderate African Americans, the slope, while positive, is not significant. That is, more liberal Republican voters do not perceive themselves to be more ideologically aligned with moderate African American Republicans than more conservative voters do. This was not the case with women, where moderate female candidates did enjoy an admittedly small congruence gain among more liberal voters in the party. Panel E is even more indicting. Whereas in the gender setting there was not a perceived difference between men and women sharing the same ideological views, in the racial setting there is among moderate candidates. African American and white moderate candidates share similar slopes but their intercepts vary significant. Republican voters across all ideology levels perceive themselves as more ideologically incongruent with African American candidates than white American candidates, even ideological moderates. These results represent some of the first evidence in this study that there may be a racial backlash for counterstereotypical African Americans in the Republican Party. It may be that a moderate policy message coupled with a liberal demographic cue generates pause among Republicans, and they do not dismiss 132

150 demographic information to the extent the theory predicted. Indeed, in this instance, the results are inconsistent with the theory, but they are unpredicted in a way that is plausible African American moderates in the Republican Party appear to have a harder time establishing ideological congruity with more liberal Republican voters than do white male moderates. They suggest an important implication for electoral politics if Republican voters do not distinguish candidates along racial lines when candidates are conservative, but they may do so when candidates are moderate, African American Republicans should be particularly incentivized to move to the ideological right in primaries. 133

151 Table 28: Perception of Ideological Congruence with White and African American Candidates by Policy Message Type (Republicans) Moderate Conservative Respondent Conservatism (0.121) (0.114)** Party Affect (0.005)** (0.003) Weak Partisan (0.245)* (0.183) Age (0.007) (0.004) Shared Race (0.612) (0.569) Female respondent (0.193) (0.138) Education (0.060) (0.047) Income (0.025)* (0.019) Latent Racism (0.107)* (0.060) White Candidate (0.623) (0.586) African American Candidate (0.218) (0.138) Constant N F R RMSE

152 Panel A: Conservative White vs. Moderate Afr. Am. Panel B: Conserv. Afr. Am. Vs. Moderate Afr. Am. Perceived Ideological Incongruence with Candidate Respondent's Conservatism Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_White Moderate_White Conservative_AfrAm Moderate_AfrAm Panel C: Conservative White vs. Moderate Afr. Am. Panel D: Moderate White vs. Conservative Afr. Am. Perceived Ideological Incongruence with Candidate Respondent's Conservatism Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_White Moderate_AfrAm Conservative_AfrAm Moderate_White Panel E: Conservative White vs. Conserv. Afr. Am. Panel F: Moderate White vs. Moderate Afr. Am Perceived Ideological Incongruence with Candidate Respondent's Conservatism Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Panel G: All Candidates Perceived Ideological Incongruence with Candidate Respondent's Conservatism Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Figure 28: Perception of Ideological Incongruence by Race and Policy Message Type (Republicans) 135

153 5.3 THE EFFECTS OF DEMOGRAPHIC CUES AND POLICY INFORMATION ON IDEOLOGICAL PERCEPTIONS: DEMOCRATS Democratic voters tend to behave in similar ways to Republicans. Policy information is far more relevant than demographic information in the course of inferring about candidate ideology. Since the results are largely analogous to those discussed in this section, I do not include a separate discussion of Democrats here. The same estimations for Democrats are available in Appendix C. The principal difference between Republicans and Democrats is that among the latter group the effects are not as robust. Slopes and predicted values of ideology/ideological congruence are all in the anticipated direction, but there are more instances of overlapping confidence intervals. Even with these minor differences, however, the results are quite consistent with Republicans, allowing me to wrap up the chapter with a discussion of the key findings that appear highly generalizable across parties. 5.4 CONCLUSIONS: POLICY INFORMATION SHAPES IDEOLOGICAL PERCEPTIONS This chapter set out to assess the impact of race and gender in a high-information environment in which demographic information was accompanied by policy messages with distinct ideological tones. According to the theory above, as specific information increases about candidates, race and gender should be increasingly irrelevant to the formation of perceptions about candidate ideology. This contention is supported by the empirical evidence. However, before summarizing and discussing the specific findings in this chapter, I must reiterate that given the 136

154 absence of demographic-based variation in candidate evaluations in a low-information environment in the previous chapter, it is possible that some of the null findings here may result from the same phenomena that influenced the null findings previously. That said, however, in this chapter a theoretical claim for not finding demographic effects is much more justified. Even when categorical cues like party and race are in conflict, ideological information provides a more relevant means for sorting candidates. Thus the conflict is effectively resolved. This is what occurs here. Differences in perceived ideology of candidates depend almost exclusively on what they say. Starting with the ANOVA tests and all the way through the regression analysis considering straight ideology as well as ideological incongruence, Republican (and Democratic) voters evaluate candidates ideology based on what is being said, and not what candidates look like. In the ANOVA analysis, for instance, I observed that significant perception differences occurred when ideological moderates were contrasted with ideological conservatives, but not when men were compared with women (African Americans). Similarly, predicted values of ideological conservatism were influenced heavily by the tone of candidate messages, but never by the demographic status of the message provider. There is also evidence that ideological perceptions depend in part on the voters themselves. As respondents level of conservatism grows, they perceive conservative candidates to be increasingly conservative and moderate candidates to be increasingly moderate. By extension, ideological congruence also depends on where voters see themselves in a unidimensional ideological space. The more ideologically liberal Republicans are, the more incongruence they feel with conservative candidates. Among more conservative partisans, the opposite occurs: they feel higher levels of congruence with conservative candidates while 137

155 perceiving moderate candidates to be more ideologically distant. The one exception to this pattern occurs with moderate African American Republicans. More liberal Republicans consider both moderate and conservative African American Republicans as similar in terms of ideological congruence, which is to say none of these candidates are considered especially proximate. In other words, while more liberal Republicans generally perceived less ideological incongruence with ideologically moderate candidates, this was not the case with African American Republicans. These candidates are not perceived to be ideologically similar with any group of Republican voters. Above I speculated that this may be due to the fact that a moderate policy message coupled with a liberal demographic cue may have been perceived as too moderate, even by more liberal Republican partisans. Yet it is interesting to note that this pattern only occurs among African American Republicans women do not struggle to achieve congruence with moderate voters. It should also be pointed out that given the relative paucity of liberal Republicans, the small number of observations at this end of the scale makes it more difficult to uncover racial effects if they do indeed exist. Nevertheless, this represents some of the first potential effects of demographic cues in this analysis of primary elections. Ultimately, the evidence here supports the theory above, but, as I have outlined, the results need to be taken with a grain of salt. The evidence suggests that race and gender do not matter (much) when candidates provide ideological information about where their political outlook lies within the field of party candidates. This is consequential. It appears that demographic cues associated with liberal stereotypes are not the burden we may presume them to be in Republican primary elections. Instead, voters focus principally on policy messages and evaluate candidates ideologies on the basis of them. 138

156 Even if voters do not respond to certain candidate characteristics, their perceptions are colored in part by their own. Indeed, a key ancillary finding in this chapter is that respondent s own ideology significantly affects how they perceive the ideology of others. While this is not a new finding in and of itself, this chapter represents the first evidence of this pattern occurring in a primary context. Voters across different levels of ideology routinely perceive different ideological gaps between themselves and different (ideological) types of candidates. This finding underscores the importance of policy messages in primary elections, for they imply that ideological sorting is an important aspect of the evaluative process among primary voters. 139

157 6.0 CHAPTER SIX: ATTITUDES TOWARDS CANDIDATES The most consistent pattern to emerge from the data so far is that demographic cues have little impact in candidate evaluation in intra-partisan primary contests. This was contrary to expectations in the first empirical chapter, but in the last one it is largely consistent with expectations: when candidates establish an ideological identity via policy messages, voters respond to the more specific information rather than categorical demographic cues. Still unaddressed, however, is the crucial question of whether counterstereotypical status can benefit female and African American Republicans. This issue is taken up in this chapter. I assess whether race, gender, and policy information interact in such a way that female and African American candidates may actually be preferred to white male candidates under certain conditions. Since demographically counterstereotypical candidates do not exist in the Democratic treatment conditions, I focus exclusively on Republicans in this analysis. 6.1 THE EFFECTS OF DEMOGRAPHIC CUES AND POLICY INFORMATION ON CANDIDATE EVALUATIONS: REPUBLICANS I begin this analysis with simple two-way ANOVA tests in which, as in the last chapter, I interact the policy factor (moderate, conservative) with the demographic factor (control, white, female, African American). Since the number of interactions created by doing this is large (28), 140

158 I focus for the moment only demographic comparisons within ideological groups (e.g. moderate candidates vs. moderate candidates). Theoretically, the difference in perceived ideology should not differ. While these tests to not explicitly account for the ideology of respondents, they provide a useful first step in comparing (1) support and (2) representation capacity across different types of candidates. Table 29 presents the ANOVA results across demographic conditions on the issue of candidate support. The results show no significant differences across groups Republican voters do not support a particular type of conservative over moderate or vice versa. As before, it is ideological information that is driving the results. There are robust main effects of ideology (F = 13.41, Prob > F = 0.000) but not for the demographic factor (F = 0.13, Prob > F = 0.923) or the interaction between the two. When specific candidates of different ideologies are contrasted, though, there is little evidence suggesting meaningful differences. This is somewhat curious, for I would expect Republicans in general to prefer conservative candidates to more moderate ones. Yet this may also be due in part to that counterstereotypical status is somewhat beneficial to female and African American candidates and therefore even moderate counterstereotypes can maintain an equal support footing with conservative white men. Alternatively, however, I have cautioned throughout, this may be an artifact of the design if demographic cues were not as stimulative as originally expected. 141

159 Table 29: Differences in Support for Candidates Across Ideological Subgroups (Republicans) Comparison Contrast Standard error T White Male (moderate) vs. Female (moderate) White Male (moderate) vs. African American (moderate) Female (moderate) vs. African American (moderate) White Male (conservative) vs. Female (conservative) White Male (conservative) vs. African American (conservative) Female (conservative) vs. African American (conservative) White Male (moderate) vs. Female (conservative) White Male (moderate) vs. African American (conservative) Female (moderate) vs. African American (conservative) White Male (conservative) vs. Female (moderate) White Male (conservative) vs. African American (moderate) Female (conservative) vs. African American (moderate) I estimate the same two-way ANOVA test for the issue of representation capacity. The results are presented in Table 30. The patterns are analogous to those in Table 29. Demographic information does not stimulate differences in perceptions about candidates abilities to represent voters interests, at least when pooled in this way, with one possible exception. There is a slightly higher level of support for female conservatives over moderate African American candidates (significant at the 0.10-level, two-tailed test), but this is the only instance of a significant interaction. The lack of significant results in these early tests suggests that if there is variation in attitudes towards candidates it occurs at a more mirco level. 142

160 Table 30: Differences in Representation Capacity of Candidates Across Ideological Subgroups (Republicans) Comparison Contrast Standard error T White Male (moderate) vs. Female (moderate) White Male (moderate) vs. African American (moderate) Female (moderate) vs. African American (moderate) White Male (conservative) vs. Female (conservative) White Male (conservative) vs. African American (conservative) Female (conservative) vs. African American (conservative) White Male (moderate) vs. Female (conservative) White Male (moderate) vs. African American (conservative) Female (moderate) vs. African American (conservative) White Male (conservative) vs. Female (moderate) White Male (conservative) vs. African American (moderate) Female (conservative) vs. African American (moderate) Gender Effects Support To explore the effects of voter characteristic on attitudes towards candidates I estimate a series of regression models as I did before and plot predicted values of these variables as a function of voter ideology. Table 31 presents a series of models predicting voters willingness to support particular candidates based on (1) voter s raw ideology and (2) the perceived ideological 143

161 congruence between voter and candidate. If the corresponding hypotheses (9a 9d) are supported by the data, I should observe that voters tend to express more support for counterstereotypical candidates than stereotypical cones at various ideological levels. As in the last chapter, I analyze female and African American candidates separately. 144

162 Table 31: Effects of Gender, Ideology, and Ideological Incongruence on Support for Candidates Moderate Conservative Respondent Conservatism (0.099) (0.056)** - Perceived Ideological Gap (0.976) (0.084)** Party Affect (0.006) (0.006) (0.004) (0.004) Weak Partisan (0.272) (0.204) (0.149)** (0.146)** Age (0.008) (0.007) (0.005) (0.005) White respondent (0.388) (0.236) (0.260) (0.245) Shared Gender (0.204) (0.182) (0.139) (0.142)* Education (0.082) (0.072) (0.056) (0.020) Income (0.028) (0.023) (0.018) (0.054) Latent Sexism (0.125) (0.102) (0.086) (0.095) Female Candidate (0.258) (0.219) (0.163) (0.172) Male Candidate (0.219) (0.198) (0.192) (0.182) Constant N F R RMSE

163 Panel A: Conservative Female vs. Moderate Female Panel B: Conservative Male vs. Moderate Male Support for Candidate Support for Candidate Respondent's Conservatism Respondent's Conservatism Conservative_Female Moderate_Female Conservative_Male Moderate_Male Panel C: Conservative Female vs. Moderate Male Panel D: Conservative Male vs. Moderate Female Support for Candidate Support for Candidate Respondent's Conservatism Conservative_Female Moderate_Male Respondent's Conservatism Conservative_Male Moderate_Female Panel E: Conservative Female vs. Conservative Male Panel F: Moderate Female vs. Moderate Male Support for Candidate Support for Candidate Respondent's Conservatism Conservative_Female Conservative_Male Respondent's Conservatism Moderate_Female Moderate_Male Panel G: All Candidates Support for Candidate Respondent's Conservatism Conservative_Female Conservative_Male Moderate_Female Moderate_Male Figure 29: Effects of Gender and Ideology on Support for Candidates 146

164 Note that the models for both respondent ideology and perceived ideological congruence are consolidated into one table. As should be expected, the coefficients for these ideological measures are negative (though shy of statistical significance) in the moderate candidate condition and positive and highly significant in the conservative candidate condition. Figure 29 presents estimates of values of candidate support across different pairs of candidates as function of voter ideology. Ideology continues to be the more powerful factor. The graphs show that when a conservative candidate is compared to a moderate candidate (Panels A to D), the slope for moderate candidates is negative and the slope for conservative candidates is positive. The most liberal of Republicans, in other words, prefer the moderate candidate to the conservative candidate and the most conservative Republicans express more support for the more conservative candidate, regardless of the gender of either candidate. A key test of Hypothesis 9, however, comes when two candidates with the same ideology are contrasted. Panels E and F present these results. It is logical that both slopes are positive since both candidates share the same ideology. Interestingly, the female candidate s intercept is higher than the conservative white male s intercept. The effect, while consistent with the expectations violations part of the theory, falls shy of statistical significance. I cannot conclude with certainty that conservative women enjoy more support than men, but the results do suggest at a minimum that parity is achieved. The same conclusion applies to moderate candidates. Here male candidates rate slightly more support at each ideological level, but the effect is clearly insignificant. On the issue of support, then, there is little evidence to suggest counterstereotypical status is an electoral asset. Voters do respond to ideology, but unexpected candidates do not enjoy an affective boost from not looking the part. 147

165 Panel A: Conservative Female vs. Moderate Female Panel B: Conservative Male vs. Moderate Male Support for Candidate Support for Candidate Perceived Ideological Incongruence Conservative_Female Moderate_Female Perceived Ideological Incongruence Conservative_Male Moderate_Male Panel C: Conservative Female vs. Moderate Male Panel D: Conservative Male vs. Moderate Female Support for Candidate Support for Candidate Perceived Ideological Incongruence Conservative_Female Moderate_Male Perceived Ideological Incongruence Conservative_Male Moderate_Female Panel E: Conservative Female vs. Conservative Male Panel F: Moderate Female vs. Moderate Male Support for Candidate Support for Candidate Perceived Ideological Incongruence Conservative_Female Conservative_Male Perceived Ideological Incongruence Moderate_Female Moderate_Male Panel G: All Candidates Support for Candidate Perceived Ideological Incongruence Conservative_Female Conservative_Male Moderate_Female Moderate_Male Figure 30: Effects of Gender and Ideological Incongruence on Support for Candidates 148

166 The results are similar when the key independent variable of interest is the perceived incongruence between voter and candidate. Figure 30 presents these predictions. Note that here the x-axis is ideological incongruence between voter and candidate. Higher values (moving left to right) indicate a larger perceived ideological gap between where voters view themselves and where they locate candidates. As should be expected, values of support decrease the more voters perceive candidates to be ideologically distant. When congruence is high, however, there are almost no demographic or ideological effects. For instance, when voters view themselves as either ideologically equivalent or one-unit away from a candidate, there is no preference for any particular demographic group or ideological group. This is clear in Panels A to D. When congruence is high (the left-hand part of the figures), differences between candidates are slight. As that incongruence increases, though, differences emerge. These distinctions are entirely ideologically based the slope for the moderate candidate regardless of gender is steep while the slope for the conservative candidate again, regardless of gender is shallower. This means that once voters perceive candidates to be about two ideological units away from themselves, they start to view moderate and conservative candidates differently. Support lingers for conservative candidates longer than moderate ones. The key, though, is that this effect is independent of gender. As above, the real test of the expectancy violation hypothesis comes when men and women with the same ideology are contrasted. This is shown in Panels E and F. Once again, when two conservatives are compared the female candidate actually enjoys higher levels of support than men, but the effect is not statistically significant. When two moderates face off, levels of support for men and women are nearly identical across the range of ideological congruence. All told, these results are in some ways inconsistent with the theory but add 149

167 empirical evidence to the conclusions reached throughout this project. Primary voters respond principally to ideological information, not gender cues. While these results should not be interpreted as evidence that sexism has been eradicated within the Republican Party, they are at least consistent with the notion that given particular party labels and ideological information, gender-based ideological stereotypes may not be as pervasive as once thought Representation Capacity The same analysis is conducted on the issue of representation capacity. Again, results are broken down by ideology and ideological congruence. As Table 32 demonstrates, among conservative political candidate there is a positive and significant increase in perceived representation capacity for female candidates (relative to the control group) but not for white male candidates. This is the first evidence that counterstereotypical status may yield some electoral gains for female Republicans. Indeed, turning to Panel E in Figure 31, we observe that female conservatives enjoy higher representation ratings than do white male candidates. As respondents grow increasingly conservative, this difference becomes statistically significant (at the 90% confidence level). This result, while only moderately robust statistically, is quite telling substantively. In this instance two candidates with the same ideology are compared and the female conservative is perceived as a better representative than the male conservative. The data provide some evidence in support of the potential for counterstereotypical status to be beneficial for female candidates, although it is far from conclusive evidence given the previous findings. 150

168 Table 32: Effects of Gender, Ideology, and Ideological Incongruence on Perceived Representation Capacity Moderate Conservative Respondent Conservatism (0.089) (0.054)** - Perceived Ideological Gap (0.074)** (0.067)** Party Affect (0.006) (0.005)** (0.004)** (0.005)** Weak Partisan (0.246) (0.183)* (0.142)** (0.142) Age (0.008) (0.006) (0.005) (0.005) White respondent (0.332) (0.304) (0.239) (0.235) Shared Gender (0.183) (0.164) (0.123) (0.126) Education (0.076) (0.067) (0.050) (0.050) Income (0.025) (0.021) (0.016) (0.017)** Latent Sexism (0.111) (0.091) (0.087) (0.094) Female Candidate (0.228) (0.203) (0.150)** (0.161)** Male Candidate (0.204) (0.188) (0.179) (0.169) Constant N F R RMSE

169 Panel A: Conservative Female vs. Moderate Female Panel B: Conservative Male vs. Moderate Male Perception of Candidate's Representation Capacity Respondent's Conservatism Perception of Candidate's Representation Capacity Respondent's Conservatism Conservative_Female Moderate_Female Conservative_Male Moderate_Male Panel C: Conservative Female vs. Moderate Male Panel D: Conservative Male vs. Moderate Female Perception of Candidate's Representation Capacity Respondent's Conservatism Conservative_Female Moderate_Male Perception of Candidate's Representation Capacity Respondent's Conservatism Conservative_Male Moderate_Female Panel E: Conservative Female vs. Conservative Male Panel F: Moderate Female vs. Moderate Male Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Respondent's Conservatism Respondent's Conservatism Conservative_Female Conservative_Male Moderate_Female Moderate_Male Panel G: All Candidates Perception of Candidate's Representation Capacity Respondent's Conservatism Conservative_Female Conservative_Male Moderate_Female Moderate_Male Figure 31: Effects of Gender and Ideology on Perceived Representation Capacity of Candidates 152

170 Panel A: Conservative Female vs. Moderate Female Panel B: Conservative Male vs. Moderate Male Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Conservative_Female Moderate_Female Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Conservative_Male Moderate_Male Panel C: Conservative Female vs. Moderate Male Panel D: Conservative Male vs. Moderate Female Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Perceived Ideological Incongruence Conservative_Female Moderate_Male Conservative_Male Moderate_Female Panel E: Conservative Female vs. Conservative Male Panel F: Moderate Female vs. Moderate Male Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Perceived Ideological Incongruence Conservative_Female Conservative_Male Moderate_Female Moderate_Male Panel G: All Candidates Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Conservative_Female Conservative_Male Moderate_Female Moderate_Male Figure 32: Effects of Gender and Ideological Incongruence on Perceived Represent. Capacity of Candidates 153

171 Figure 32 presents the results of ideological congruence and perceptions of representation capacity. The results suggest caution must be taken when inferring about expectancy violations because Panel E in this figure shows that while conservative women continue to enjoy higher levels of support than conservative men across ideological values, the effect here does not quite achieve statistical significance. The rest of the figure closely mirrors the results from the support variable. When ideological incongruence is low, voters tend not to distinguish among different types of candidates. As incongruence grows, moderate candidates of both genders see a sharp decline in perceived representation capacity while conservative candidates of both genders see a shallower decline. The results affirm that it is not merely ideology, but ideological congruence, that determines attitudes towards candidates. The results show that when congruence is high, other information even ideological policy information is not particularly compelling. Simply put, voters support and view as good representatives the candidates with whom they are most ideologically aligned Racial Effects Support The same analysis is here applied to race instead of gender. Although there are some similar patterns across the counterstereotypical groups, there are some distinct differences, as well. Consider first Table 33 and Figure 33. These results focus on the effects of race and respondent ideology on support for white versus African American candidates. In general, the candidates slopes are all in the expected direction (e.g. as voters grow more conservative, they express more support for conservative candidates). In Panels A to D, it is clear that towards the conservative 154

172 end of the ideological spectrum Republican voters clearly distinguish moderate from conservative candidates. As with women, more conservative Republicans express more support for the conservative candidate, regardless of race. Yet it is interesting to note that, unlike with women, there is not parallel process occurring at the other end of the spectrum. That is, in the gender analysis, more liberal Republicans expressed more support for the moderate candidate over the conservative one (see Panels A to D in Figure 31). Yet in the racial analysis this does not occur while the moderate candidate is rated higher than the conservative candidate, as expected, the difference is not statistically significant. It should be noted that these results could be artifactual owing to the relatively few number of liberal Republicans in the sample (and the population). Yet the same was true for the gender analysis, and some important distinctions emerged there. Consequently, another interpretation worth considering is that Republican voters may be somewhat more comfortable with moderate female party candidates than moderate African American party candidates. Of the two counterstereotypes, women are far more common in the Republican Party than African Americans. As a result, the considerable variance in the case of moderate African American Republicans may partially be due to some Republicans hesitating to support unconventional candidates, even when those candidates ideologies align well with the voters. As before, however, an important test of the expectancy violations hypothesis comes in Panels E and F in which demographic status is constant across different ideological types of candidates. Here there is some evidence supporting the theoretical expectation that within ideological subgroups counterstereotypical candidates may gain some affective boost. In Panel E, across nearly every level of ideology, the conservative African American candidate enjoys significantly higher levels of support than does the equally conservative white male candidate. 155

173 Since ideology is effectively held constant, the only explanation is that the counterstereotypical African American candidate benefited simply by being a conservative African American. Table 33: Effects of Race, Ideology, and Ideological Incongruence on Support for Candidates Moderate Conservative Respondent Conservatism (0.109) (0.073)** - Perceived Ideological Gap (0.057) (0.092)** Party Affect (0.006)** (0.003)** (0.005) (0.005) Weak Partisan (0.273) (0.138) (0.180)** (0.172)** Age (0.008) (0.005) (0.006) (0.006) Shared Race (0.617) (0.240) (0.595) (0.467) Female Respondent (0.210) (0.146) (0.166) (0.167) Education (0.067) (0.052) (0.059) (0.057) Income (0.025)** (0.017) (0.020)* (0.021) Latent Racism (0.104) (0.068) (0.075) (0.073) White Candidate (0.628) (0.246) (0.634) (0.509)* African American Candidate (0.231) (0.158) (0.184) (0.188)* Constant N F R RMSE

174 Turning to moderate candidates, the same expectancy violation hypothesis is not supported. Here white and African American candidates follow similar positive tracks across ideology but there is never a statistical distinction made between them by voters at any ideological level. Coupled with the results from the gender analysis, the evidence overall suggests that violating expectations may yield some benefit to counterstereotypical candidates but only if they are ideologically conservative. When they are ideologically moderate they tend to be rated similarly to white male candidates. That is, counterstereotypical candidates appear neither to lose nor gain from ideological moderation, but they do potentially gain from being conservative. 157

175 Panel A: Conservative White vs. Moderate White Panel B: Conserv. Afr. Am. Vs. Moderate Afr. Am. Support for Candidate Support for Candidate Respondent's Conservatism Respondent's Conservatism Conservative_White Moderate_White Conservative_AfrAm Moderate_AfrAm Panel C: Conservative White vs. Moderate Afr. Am. Panel D: Moderate White vs. Conservative Afr. Am. Support for Candidate Support for Candidate Respondent's Conservatism Respondent's Conservatism Conservative_White Moderate_AfrAm Moderate_White Conservative_AfrAm Panel E: Conservative White vs. Conserv. Afr. Am. Panel E: Moderate White vs. Moderate Afr. Am Support for Candidate Respondent's Conservatism Conservative_White Conservative_AfrAm Support for Candidate Respondent's Conservatism Moderate_White Moderate_AfrAm Panel G: All Candidates Support for Candidate Respondent's Conservatism Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Figure 33: Effects of Race and Ideology on Support for Candidates 158

176 Panel A: Conservative White vs. Moderate White Panel B: Conserv. Afr. Am. Vs. Moderate Afr. Am. Support for Candidate Support for Candidate Perceived Ideological Incongruence Perceived Ideological Incongruence Conservative_White Moderate_White Conservative_AfrAm Moderate_AfrAm Panel C: Conservative White vs. Moderate Afr. Am. Panel D: Moderate White vs. Conservative Afr. Am. Support for Candidate Support for Candidate Perceived Ideological Incongruence Conservative_White Moderate_AfrAm Perceived Ideological Incongruence Moderate_White Conservative_AfrAm Panel E: Conservative White vs. Conserv. Afr. Am. Panel E: Moderate White vs. Moderate Afr. Am Support for Candidate Perceived Ideological Incongruence Support for Candidate Perceived Ideological Incongruence Moderate_White Moderate_AfrAm Conservative_White Conservative_AfrAm Panel: All Candidates Support for Candidate Perceived Ideological Incongruence Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Figure 34: Effects of Race and Ideological Congruence on Support for Candidates 159

177 When it comes to ideological congruence, the racial analysis bears similar results as the gender tests above. Figure 34 shows that, as should be expected, the slopes for both moderate and conservative candidates are negative in all panels. As ideological incongruence increases, voters are less likely to support any candidate. In Panels A and C the slopes decline at different rates but they are never statistically distinguishable from each other. In Panels B and D, however, as ideological incongruence increases Republicans begin to express different levels of support for moderate and conservative candidates. In both of these panels the conservative candidate is African American, and is contrasted with a moderate African American and a moderate white candidate in Panels B and D, respectively. It follows that the conservative African American candidate enjoys a (marginally) significantly statistical edge over moderate candidates while stereotypical white male conservatives do not. The results are hardly conclusive, but they are consistent with the previous results suggesting the conservative African American Republican candidates can benefit from their counterstereotypical status. There are limits to this conclusion, however. Panels E and F in Figure 34, for instance, indicate that conservative candidates of either race lose support as ideological incongruence increases. That is, African American conservatives do not experience a more shallow drop in support than their white male counterparts. Nor do African American moderates. Take with the raw ideology plots in Figure 33, the evidence here suggests that counterstereotypical status can be beneficial when voters are comparing ideologically like-minded candidates. However, ideological congruence is more important that expectancy violations. If voters perceive any candidates to be relatively distant from themselves ideologically, they will express less support for those candidates regardless of race. 160

178 Representation Capacity The last empirical section considers the effects of race and ideology on how voters perceive white and African American candidates as potential representatives. Table 34 and Figures 35 and 36 present the results. Panels A to D in Figure 35 display a similar pattern as emerged in the gender analysis above. Differences in perceived representation capacity are a function of candidate ideology, not demographic status. In each of these four panels, conservative candidates enjoy a positive slope as respondents grow more conservative. Moderate candidates, on the other hand, enjoy significantly higher representation ratings among more liberal Republicans, but see a steady decline as respondents conservatism grows. These effects are similar across all racial combinations of candidates, and are entirely consistent with the notion that ideological information trumps demographic considerations in candidate evaluations. The test of the expectancy violations hypothesis in Panels E and F provide little evidence of an effect. As with women, in Panel E conservative African American candidates are rated as better representatives than white male conservatives, but the effect does not achieve statistical significance. Moreover, in Panel F the moderate white male is rated slightly higher than the moderate African American, although again the effect is not close to being a significant one. Once again the evidence for a pure expectancy violations gain for counterstereotypical candidates is mixed. There are some instances in which it appears to play a role, but other times the effects are small, insignificant, or do not appear at all. 161

179 Table 34: Effects of Race, Ideology, and Ideological Incongruence on Perceived Representation Capacity Moderate Conservative Respondent Conservatism (0.100) (0.074)** - Perceived Ideological Gap (0.082)** (0.081)** Party Affect (0.006)** (0.005)** (0.006)* (0.006)* Weak Partisan (0.257)** (0.173)** (0.167)** (0.170(** Age (0.009) (0.007)** (0.005) (0.005) Shared Race (0.541) (0545) (0.571) (0.494) Female respondent (0.200) (0.178) (0.160) (0.164) Education (0.085) (0.074) (0.059) (0.059) Income (0.030) (0.027) (0.017) (0.018)** Latent Racism (0.109) (0.088) (0.069) (0.070) White Candidate (0.560) (0.0553) (0.617) (0.546)* African American Candidate (0.240) (0.205) (0.180) (0.188) Constant N F R RMSE

180 Panel A: Conservative White vs. Moderate White Panel B: Conserv. Afr. Am. Vs. Moderate Afr. Am. Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Respondent's Conservatism Respondent's Conservatism Conservative_White Moderate_White Conservative_AfrAm Moderate_AfrAm Panel C: Conservative White vs. Moderate Afr. Am. Panel D: Moderate White vs. Conservative Afr. Am. Perception of Candidate's Representation Capacity Respondent's Conservatism Perception of Candidate's Representation Capacity Respondent's Conservatism Moderate_White Conservative_AfrAm Conservative_White Moderate_AfrAm Panel E: Conservative White vs. Conserv. Afr. Am. Panel E: Moderate White vs. Moderate Afr. Am Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Respondent's Conservatism Respondent's Conservatism Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Panel: All Candidates Perception of Candidate's Representation Capacity Respondent's Conservatism Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Figure 35: Effects of Race and Ideology on Perceived Representation Capacity of Candidates 163

181 Panel A: Conservative White vs. Moderate White Panel B: Conserv. Afr. Am. Vs. Moderate Afr. Am. Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Perceived Ideological Incongruence Conservative_White Moderate_White Conservative_AfrAm Moderate_AfrAm Panel C: Conservative White vs. Moderate Afr. Am. Panel D: Moderate White vs. Conservative Afr. Am. Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Perceived Ideological Incongruence Conservative_White Moderate_AfrAm Moderate_White Conservative_AfrAm Panel E: Conservative White vs. Conserv. Afr. Am. Panel E: Moderate White vs. Moderate Afr. Am Perception of Candidate's Representation Capacity Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Perceived Ideological Incongruence Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Panel: All Candidates Perception of Candidate's Representation Capacity Perceived Ideological Incongruence Conservative_White Conservative_AfrAm Moderate_White Moderate_AfrAm Figure 36: Effects of Race and Ideological Incongruence on Perceived Represent. Capacity of Candidates 164

182 Finally, Figure 36 presents the effects of race and ideological congruence. As expected, and consistent with previous findings, the slopes for moderate and conservative candidates are negative in all panels, further affirming that voters recognize ideological incongruence with candidates are modify their evaluations of them accordingly. Moderates of both races see a sharper decline in perceived representation capacity relative to conservatives of both races, but only when ideological incongruence is extremely high do these differences yield significantly different predictions. This last figure contributes additional evidence to support the claim that when ideological information is present, candidates demographic status is largely disregarded by voters. If valid, this is a normatively encouraging pattern. 6.2 CONCLUSIONS: SOME EVIDENCE OF EXPECTANCY VIOLATIONS This chapter sought to extend the ideological perceptions analysis in Chapter 5 to other important attitudes about candidates. It is here I anticipated that expectancy violations may lead to a significant electoral boost for female and African American candidates over stereotypical white male candidates in Republican primaries. There is indeed some evidence of this, but it is not overwhelming, nor is it distributed evenly across all candidate types. On the issue of candidate support, there is a modest increase for conservative women (and African American men) over conservative men across Republican voters. Given the experimental design, this effect is consistent with an expectancy violations model unexpected candidates within ideological subgroups do stimulate some positive affect relative to conventional candidates. In some cases, as in conservative African American candidates relative to conservative white male candidates, 165

183 the effect is relative substantive around and even over one point on a seven point scale. Some Republicans also perceive conservative women as somewhat better representatives than conservative white males. However, the effects are not always this pronounced. The increased difference in support for women over men, for instance, does not quite achieve statistical significance. Nor do Republican voters perceive differences in representation capacity between conservative white men and conservative African American men. Most importantly, demographic-based differences in support and representation capacity only emerge among conservative candidates. When moderate candidates are contrasted, race and gender do not generate positive affect, perhaps because moderate policy information works in conjunction with a liberal demographic cue to make Republican voters a bit wary. One interesting finding in this analysis more generally is that demographic information does interact with policy information to generate (some) significant differences across candidate groups. So far the empirical work has suggested that demographic cues simply do not matter. The results in this chapter qualify that pattern race and gender certainly appear to be less powerful cues when applied in a high-information context, but there are several instances discussed above where they do influence evaluations, for better in some cases, for worse in others. In the next and final chapter I synthesize the findings of the two studies here and discuss their implication for primary politics in America. 166

184 7.0 CHAPTER SEVEN: DISCUSSION AND CONCLUSION This project represents a foray into information use in primary elections and how voters utilize cues entirely within a friendly but complex partisan context. Specifically, I focused on the role of race and gender within party primaries at a time when the ever-shifting demographic make-up of the United States is beginning to have serious implications for the long-term viability of at least one major political party. The Republican Party in particular has struggled to diversify its constituency and is actively pursuing strategies to recruit more women and people of color to run for office under the party banner. While well-intentioned, these efforts, I argue, may not resolve the issue if voters are not also willing to support these candidates in party primary elections where their counterstereotypical status may in theory pose problems in winning over the primary electorate. The rest of the project is designed to help advance our understanding of how and when this support occurs. I develop a theory in which I argue that candidate success in primaries is a function of both voter-centric and candidate-centric attributes. Since party labels do fulfill their usual role as go-to heuristic for voters selecting a candidate, and since the electorate is more ideologically homogenous during primaries, race, gender, and the voters self-identification within parties should be more central to decision-making than they would be in general elections. Given the ideological signals embedded within demographic cues, I argue that when primary voters know little about a candidate other than his or her race and gender as well as status as a co-partisan, 167

185 they will apply sort primary options and select a most-preferred option. The second element of the theory proposes that counterstereotypical candidates presenting conflicting information cues say, an African American Republican can signal to voters which piece of information is more relevant to the decision-making process through policy messages contained in their rhetoric. Candidates can, in effect, resolve the cognitive conflict generated by their counterstereotypical status. Two survey experiments explore these dynamics. The rest of this chapter contains four sections: First, I summarize key findings from the studies explored in detail above. Second, I return to and revise the original Republican dilemma introduced in Chapter 1, proposing that conventional wisdom about counterstereotypical candidates, while not inaccurate, is incomplete. Third, I address the implications these results have for the Republican Party moving forward in the 2016 election and beyond. Finally, I discuss how these results inform a research agenda that continues to explore the nature of information and decision-making in primary elections. 7.1 KEY FINDINGS This project set out to more clearly reveal the interplay between demographic, partisan, and policy identities in primary elections. Thematically, perhaps the most significant general contribution of this work is to illustrate how context can dramatically affect the nature of influence of information. I uncover evidence suggesting that the relative impact of different types of information depends heavily on the context of the interaction between voter and candidate. A candidate s demographic status matters quite a bit in certain conditions, but is largely irrelevant in others. Voters perceptions and evaluations of candidates are often a 168

186 function of their own ideological position and how much they see themselves as ideologically similar or distinct from candidates. I consider in turn how variance in voters and candidates, respectively, affects the evaluative process Voters One of the central arguments I put forth above is that in primary elections the ideological identity of voters i.e. where they see themselves within their parties is central to how they use information to evaluate primary candidates. Partisans indeed differ considerably even within parties. Even when race and gender proved to have minimal effects on candidate evaluations, a consistent pattern in the data is that voters imposed their own ideological viewpoints on their perceptions of candidates. This phenomenon manifested itself in predictable ways. For instance, more conservative Republicans saw more conservative Republican candidates as increasingly ideologically proximate while also perceiving more moderate candidates as more ideologically distant. While not a direct interest in this project, these results underscore the urgency with which we must (better) understand how information is used to sort primary candidates. Indeed, this discussion motivates a key conclusion/qualification of the theory proposed above. While voters appear to use race and gender in some circumstances, the results of the experiments above appear to be more a function of ideological signals than pure expectancy violations. There is minimal evidence that merely being counterstereotypical is advantageous the first study shows no gains for these candidates in low-information environments. Instead, counterstereotypical status is only advantageous when coupled with conservative policy information. When this occurs, it (1) levels the playing field among strong Republican voters while (2) leading to gains for counterstereotypical candidates (relative to stereotypical ones) 169

187 among weak partisans. Overall, however, for a pure expectancy violations theory to hold here, we might expect to see all Republicans or strong Republicans alone prefer the unexpected candidate this is not the case. The effects here are driven largely by ideological inference as a function of policy message Candidates The information candidates provide to primary voters also plays a significant role in determining which candidates will succeed and the circumstances under which they do. Information matters more under some conditions than others. Notably, in the low-information environment simulated in the first study, I find little evidence that demographic traits have much of an effect on how voters evaluate candidates. Instead, it appears that when voters are asked to evaluate copartisans, their evaluations depend much more on party affect than they do on the candidates. Counterstereotypical candidates do not gain among weak Republican partisans, nor do they suffer among strong Republican partisans. This is encouraging news insofar as it suggests that counterstereotypical candidates will not simply be written off as too liberal for Republican voters. Again, however, this is also a possible artifact of the experimental design and weak treatment stimuli. Design concerns aside, though, the results allow me to more forcefully conclude that what matters for candidates is the messages they furnish to voters. The nature of what candidates say directly affects how demographic cues will be incorporated into voter decisionmaking often by rendering them irrelevant. Conservative (liberal) voters tend to express more favorable impression of conservative (liberal) candidates, regardless of race and gender. The results, then, suggest that counterstereotypical African American or female Republicans may not 170

188 actually struggle with demographic stereotypes as much as once thought. Among Republicans, this is especially the case when candidates are ideologically conservative. In the next section, however, I discuss some important qualifications to this statement, and synthesize the results of this project into the larger discussion about demographic cues and party politics. 7.2 REVISING THE REPUBUBLCAN DILEMMA A critical premise of this entire project was what we may call the dilemma of the female (African American) Republican. These politicians are unusual in that they present multiple pieces of information that are not consistent with ideological stereotypes. In other words, these candidates simultaneous provide a liberal signal as well as a conservative one. Voters, consequently, may struggle to disentangle conflicting information cues and draw conclusions about these counterstereotypical candidates. As Mcdermott (1997) puts it: because a Republican woman provides voters with two competing [categorical] cues, [ ] voters may not know which cue to give more weight (278). Furthermore, it is well-established that uncertainty corresponds with lower levels of support for candidates (Alvarez 1997). The result is a dilemma for counterstereotypical candidates: voters, independent of any potential latent sexism or racism, may fail to support female or African American candidates because they are not sure what these candidates stand for. The results of this project suggest a revision of the dilemma is necessary moving forward. It is not so much that the dilemma is wrong or illusory, but simply that it is incomplete. Scholars thus far have done little to assess how voters respond to information that can break cue ties presented by counterstereotypical candidates. In this project I have taken steps to begin to 171

189 do so, and the results imply that the conventional wisdom about counterstereotypical candidates is inadequate. It requires more nuance. Indeed, many of the findings above might be taken initially to suggest that counterstereotypical candidates are not in as much electoral danger as we might have suspected after all, in most cases they even the playing field, and even reverse it among certain voters. Yet the story is not this simple. The results also illustrate the potential additional challenges these candidates face, and here I seek to revise the dilemma of the counterstereotypical Republicans to more accurately portray political reality for these candidates. This new dilemma, I argue, is comprised of three parts. Counterstereotypical candidates must hold conservative views. One of the key findings above was that counterstereotypical Republican candidates succeed in leveling the playing field and even reversing it in party primary elections. More ideologically extreme (conservative or liberal) voters heavily favor ideologically extreme candidates, and moderate partisans prefer conservative women and African Americans to conventional white men. Yet the reality is that, on average, women and persons of color tend to be more liberal than white men in both parties. As Koch (2000, 426) concludes when assessing the political views of male and female congressional candidates, the [actual] ideology scores of the women candidates [are] more liberal than those of their male counterparts [ ]. Thus, citizens' generalization that women candidates are more liberal than male candidates [is] in fact true. In this project, the findings underscore a potentially distressing reality counterstereotypical candidates, who are often more ideologically moderate, really only gain an electoral edge if they are conservative. Moreover, moderate counterstereotypical candidates did not always have an advantage over conservative candidates even among ideologically moderate voters. Recall that even though ideologically moderate voters perceived themselves as 172

190 ideologically aligned with moderate candidates generally, this did not hold true for moderate African American Republicans. Together, these findings may help explain why there are relatively few counterstereotypical Republicans (assuming they campaign on sincere policy views) running for office. 42 They are also consistent with the observation that many counterstereotypical Republican candidates who achieve even moderate prominence in American politics are quite conservative these are the candidates have sufficient credentials to appeal to the Republican base. Messages must be received in order to affect evaluations. This project was not concerned with the nature of the distribution of information across the primary electorate all respondents in this study were presented with complete information about the candidates. This represents a legitimate external validity concern and an important qualification when extrapolating to party politics outside of this study. Yet even though information is held constant here, in practical political terms these conclusions suggest that the most important goal for all primary candidates is to get their policy/ideological profile out for voters to hear. The structure of primary elections in the United States, however, may hinder these efforts. At the presidential level, the sheer number of candidates struggling to be heard makes it difficult for any one voice to rise above the rest short of massive advertising efforts (and the massive costs that go with them). At the congressional level and in even lower levels of office, the struggle may be even more challenging, for voters pay even less attention to these races. While this general challenge arises for all candidates, it may be especially pronounced for counterstereotypical candidates who may face additional struggles in gaining party nomination. For instance, Lawless and Pearson (2008) conclude that women are more likely to face primary challengers than men. 42 Granted, much of this is due to the ideological distribution of female and African Americans in general. These groups tend not to be conservative and/or Republicans. 173

191 Shattered ceilings may hurt general election chances. I have committed steadfastly to analyzing primary elections exclusively in this project, but clearly we must focus somewhat on what these results mean for the general elections that follow. Unfortunately, the findings here do not paint a particularly rosy picture. We observe that female and African American candidates benefit most when they are conservative, and not simply counterstereotypical. This suggests, however, that to the extent that these candidates can successfully secure their party s nomination for the fall, the result is a party candidate potentially out of step with the general election electorate to whom s/he must now appeal. Political pundits and strategists alike often discuss the need for a candidate to move to the middle once the primary election is over in order to try and win over moderate voters in addition to the partisans who will now almost certainly vote for him. If winning the party s nomination requires counterstereotypical candidates to emphasize especially conservative credentials, it may come back to hurt them in the fall. Indeed, this may be one of the factors contributing to the general election losses by candidates like far rightleaning Sharon Angle in 2010, who lost to a very unpopular incumbent. All told, the electoral picture painted by these results represents a double-edged sword. On the one hand, there is evidence that demographic information is in many cases not particularly central to candidate evaluations in party primaries. The first survey experiment yielded evidence that voters rely much more on partisanship than race or gender; knowledge that a particular candidate shares your party identification rendered demographic cues irrelevant when inferring about candidates. The second study found that policy information is also a strong motivating factor in candidate evaluations. Strong partisans in particular care much more about what candidates say than what they look like. These results are encouraging insofar as debunking or at least deflating the proposition that women and minorities will struggle to win 174

192 in Republican party contests. On the other hand, demographic cues still matter, and they are not always electoral assets. Counterstereotypical candidates in the Republican Party tend to be ideological moderates, yet this project reveals that moderate voters those whom we should expect to be especially attracted to moderate candidates actually prefer moderate stereotypical candidates to moderate counterstereotypical candidates. The conflicting information cues presented by female/african American moderates lead to ambivalence between moderate counterstereotypical candidates and conservative ones. Coupled with other considerations message dissemination, the relative scarcity of conservative counterstereotypical Republicans, etc. and the picture is not quite as optimistic as a simple reading of the experimental results might imply. 7.3 WHAT ARE THE ADVANTAGES TO BEING COUNTERSTEREOTYPICAL? Given the discussion above, it is worth briefly revisiting precisely when counterstereotypical candidate do and do not gain an electoral advantage over their white male counterparts. There are two instances in which this edge is gained. The first is purely ideological: when counterstereotypical conservatives run against stereotypical moderates in Republican primaries, Republican voters tend to prefer the conservative counterstereotypical candidate. This is a logical conclusion given the overwhelming power of ideological information demonstrated above. Yet this finding is not trivial, for it demonstrates that voters do not hesitate to support unconventional candidates when they have policy credentials voters prefer. The second way counterstereotypical candidates may capitalize on their status is affectbased, not information-based. Above I documented some evidence that when conservative 175

193 counterstereotypical candidates (e.g. female or African American Republicans) are running against stereotypical (white male) candidates, there are some instances when voters express more favorability for the unconventional candidate. In other words, they gain an electoral edge by virtue of simply being counterstereotypical (since all information is constant). These effects are far less pronounced than the information-based gains, but they exist. Moreover, if the treatment stimuli were selected to bias the findings against demographic-based findings, the emergence of racial and gender effects may actually be more pronounced outside of this experiment. Additional research using additional/alternative stimuli will help clarify this issue. 7.4 IMPLICATIONS FOR REPUBLICAN PARTY MOVING FORWARD It is worth taking a moment to consider the results of this project in a larger political context. What do the findings above mean for the Republican Party? As the revised dilemma suggests, the results are simultaneously heartening as well as discouraging. Perhaps the most important observation, and one inconsistent with much of the research on demographic cues, is that counterstereotypical status does not tend to be a detriment to primary candidates. In low information environments, no one demographic type of candidate is preferable to another. As policy information is introduced, counterstereotypical candidates may even gain an advantage over conventional candidates when they provide the right policy information. In short, conservative counterstereotypical Republicans have the potential to do compete in Republican primaries. At the same time, we saw that moderate counterstereotypes do not enjoy the same advantages. Moderate African American candidates struggled to appeal (in some cases) even 176

194 with moderate Republican voters. Given that most counterstereotypical Republicans in the population do tend to be more moderate, this means that the potential gains African American or female Republicans may enjoy are limited to relatively few African American or female politicians the conservative ones that come to mind when we are asked to identify prominent non-white or non-male Republicans. Indeed, this is the reason that the counterstereotypical Republicans we do see tend to be quite conservative. At the same time, other considerations may be less important than we might theoretically expect. Consider primary types as an example. We might predict, a priori, that open primary states may be better for counterstereotypical Republicans since moderate independent voters can participate. Yet the findings here suggest that moderate counterstereotypical Republicans do not really have that much to gain by running in states with high concentrations of moderate voters participating. Of course, we cannot read too much into the results of a single project. These studies have limitations like all others, and we cannot conclusively determine here if counterstereotypical will always do better under one system or another. Still, the results underscore the need for the party to focus not just on descriptive represented of historically underrepresented constituencies, but also to think strategically about where and how these candidates can succeed in winning the party s nomination for the fall election. The Party should also consider the nature of information dissemination in primary elections. The revised dilemma above concludes based on the research that counterstereotypical candidates need to be heard moreso than do stereotypical ones. When messages are received, they have an effect, but we observed above that evaluations of white men do not really change as a function of policy information. They do when it is a counterstereotypical candidate speaking. Therefore the party may want to review rules how, for instance, it decides which candidates are 177

195 invited to participate in primary debates. Presumably, counterstereotypical candidates may benefit from this kind of exposure far more than conventional candidates do. If the party is truly committed to diversifying its candidate pool, it may even consider taking steps to actively promote the ability of counterstereotypical candidates to connect with primary voters. 43 Finally, it is worth taking a moment to consider recent Republican elections and how well the results of this study align with them. In many ways, the findings here reflect electoral reality. The most prominent counterstereotypical Republican candidates to emerge on the national scene Sarah Palin, Michelle Bachman, Herman Cain, and Ben Carson, to name a few are united by strong conservative credentials. None have won their party s nomination for president, but it is interesting to note that some of the most experienced moderate counterstereotypical Republicans Susan Collins and Olympia Snowe, for instance have not even attempted to do so. Moderate and libertarian candidates like Gary Johnson and Ron Paul struggled to gain traction at all, despite being stereotypical Republican candidates. Clearly, these anecdotal observations are just that, and cannot be taken as irrefutable evidence of how primary voters treat candidates more generally. Still, the results of this project are consistent with the observation that prominent counterstereotypical Republicans are almost exclusively of the conservative variety, and even then they have had mixed success. 43 This, of course, would itself be a counterstereotypical move given the Party s resistance to affirmative-action type policy positions. 178

196 7.5 CONTRIBUTIONS TO THE LITERATURE, LIMITATIONS, AND NEXT STEPS I conclude the chapter with a brief summary of the contributions this project makes to the literature as well as few additional qualifications and a discussion of how to advance this line of research. The most important contribution I make is to explicitly analyze the relationship between demographic and partisan information. Most research on demographic cues ignores partisan labels, and the few that integrate the two fail to generate consistent results. Critically, in my view, none of this work looked at party primaries where voters are asked to evaluate candidates seeking a nomination and not simply an office. However, this work is far from the last word on the issue of demographic cues in primary elections. Since I am principally concerned with better understanding how voters evaluate candidates, this project does not directly test how those perceptions affect specific and important behaviors like the decision to vote on Election Day or whether certain types of candidates motivate voters to seek additional information about them. I focused on manipulating key factors of interest demographic cues and policy messages but there are many other relevant issues that are required to paint a more complete picture of electoral behavior in primary elections. In the course of crafting a parsimonious experimental design, a number of important elements were necessarily left to future endeavors. Below I highlight a few of the most important avenues of additional work Gendered and Racialized Policy Domains In the course of developing policy messages for this study, I intentionally avoided policy domains known to have high degrees of association with particular races or genders. This was done in order to better ensure that demographics effects were indeed the result of basic 179

197 demographic differences among candidates and not due to specific issues those candidates discussed. Yet women and candidates of color are in a unique position to address and advocate issues especially relevant to the demographic groups whom they descriptively represent. Indeed, as discussed above, voters hold stereotypes about politicians policy competence, with men and women, black and white individuals perceived as being especially skillful or inept in different policy areas. Future work on race and gender in party primaries can look at not only the ideological tone of policy messages as I do here, but more acutely assess the role these sorts of policy domains play in the evaluative process. This is especially relevant given that there were some occasions in the experiments above where African American and female candidates did not generate similar reactions among voters Viability, Strategic Voting, and Multi-Candidate Contests This project is also limited in that other unique features of the primary election process are not explicitly addressed (or replicated in the experiments above). For instance, I do not simulate multi-candidate environments voters encounter in actual elections. In this project, voters only evaluate a single candidate within their party, but elections (usually) require voters to select from multiple candidates, especially in primary elections. I have several projects in the development stage that will expand on this work by asking individuals to actually select from multiple candidate options. For instance, borrowing from reference dependence theory, one such experiment will manipulate information not about a particular candidate, but instead who is running against that candidate in a primary election. I expect that perceptions of one electoral choice are shaped not merely by her own attributes, but by those of her challenger. 180

198 This project also does not consider the role for strategic voting or the concept of viability more generally. Primary elections are the first in a two-stage process, and it is not unreasonable to presume that (some) voters may make decision in the first stage that are not independent of the second stage. That is, a primary vote choice may be made with an eye towards the subsequent election. This sort of strategic voting is a central interest of mine and I already have some work under way to begin to explore this concept. For example, a colleague and I have conducted an experiment in which we attempt to identify the conditions under which a voter will defect from a sincere primary vote choice to a strategic one. Initial results suggest that some voters are indeed willing to abandon their most preferred candidate for one that has a better chance of winning the fall general election. These findings underscore the importance of incorporating viability into future endeavors related to primary elections. Race and gender offer additional opportunities to explore the strategic voting, particularly in Republican primaries as the party actively seeks to boost the presence of non-white, non-male politicians within its ranks. These considerations may also yield important insights into so-called downstream effects. For instance, scholars have noted that some candidates can attract voters from the other party in general elections. Democratic women, for example, may siphon off some Republican women in fall elections. There is additional work to do exploring how primary choices by voters affect general elections that follow spring nomination contests. For instance, latent resentment towards women or African Americans or other groups may manifest itself not in vote choice, but in turnout, with some voters being especially likely or less likely to show up at the polls when their party has nominated a particular candidate. Additionally, do party leaders devote extra time and attention to counterstereotypical candidates who advance through primary elections in the fall? 181

199 Ultimately, while the project contributes to our understanding of how the Republican Party may or may not succeed in diversifying its voting base, primary victories are only the first step on the path to increased representation Reconciling Other Forms of Information Conflict A third line of work I plan to pursue relates loosely to the concept of information conflict and how voters reconcile cognitive inconsistencies. This sort of conflict extends well beyond demographic cues. I plan to consider issues like how partisan voters deal with candidates who share some but not all conventional party policy stances. To what extent can politicians get away with not towing the party line, or engage in what we may call policy exception making? In addition, I intend to contribute to the burgeoning line of research assessing how voters respond to bipartisanship and the consequences it has for policy evaluation. In an age of increased polarization in American politics, these sorts of questions are particularly relevant to understanding how voters conceive of public policy and, ultimately, how elected officials make it. Counterstereotypical messages and concepts, as they become less common (see, e.g. the decline of Blue Dog Democrats), also become potentially more influential when they emerge Clarifying Information-Based and Affective-Based Models of Candidate Evaluation Students of political behavior have approached the question from a variety of theoretical perspectives. A social cognition approach, for instance, emphasizes affect and shared identity between voter and candidate. A Bayesian, or information-based approach emphasizes the role of new information in compelling individuals to update previously held belief. In the theory above, 182

200 I principally employ the latter approach, but argue that conditional on certain information being available, affect may also be generated via expectancy violations. Empirically, the two approaches yield similar theoretical as well as empirical expectations, yet there remains a tension between the two schools. Rather than relying on one model or the other, future work must seek to clarify the relationship between affective and informational processes, not only in primary politics, but in American politics more generally. Here I offer one way in which the two traditions may simultaneously apply to behavioral processes, but this project represent a very preliminary first step in this larger endeavor. 183

201 APPENDIX A SURVEY QUESTIONS Study 1: Candidate Evaluation and Party Subgroups Part 1: Eligibility Question Block Note: not all questions in the eligibility block directly concern eligibility. Questions 1, 2, and 5 are the pertinent queries. The others are included to minimize the likelihood that an ineligible respondent can go back and change an answer in order to make herself eligible after learning that she is not. These first few questions will be used to determine your eligibility for this survey. Please answer them honestly. Q1: What is your age in years? A1: Open-ended: Respondent inputs age Note: If respondent is under 18 years of age, the survey will end Q2: Are you a citizen of the United States? A2: Respondent selects yes or no Note: if respondent is not a citizen of the United States, the survey will end Q3: In what state do you currently reside? A3: Respondents select state of residence from dropdown menu 184

202 Q4: Which of the following best describes your religion? A4: Respondents select religion from dropdown menu (Baptist any denomination; Protestant (e.g. Methodist, Lutheran, Presbyterian, Episcopal); Catholic; Mormon; Jewish; Muslim; Hindu; Buddhist; Pentecostal; Eastern Orthodox; Other-Christian; Atheist; Agnostic) Q5: Generally speaking, do you usually think of yourself as a Republican, a Democrat, an independent or what? A5: Respondent selects Republican, Democrat, Independent, Something else Note: If respondent is neither Republican nor Democrat the survey will end Q6: Are you (Check all that apply): A6: Respondent select one or more buttons: Married, Divorced, Widowed, Separated, Single, Single parent Part 2: Interest and Media Q7: Generally speaking, how interested are you in what s going on in government and politics? A7: Respondents will select one of the following buttons: Extremely interested, Very interested, Moderately interested, Slightly interested, or Not interested at all Q8: During a typical week, about how many hours per week do you spending watching political media? A8: Open-ended: Respondent inputs number Part 3: Candidate evaluation (including experimental conditions) Note: Respondents will be randomly assigned to learn about one of four fictional political candidates within their own party. Four candidates (A white male, a white female, a black male, and a candidate with no demographic information) will provide identical policy messages. All Republican respondents, then, will be presented with one of the variations of Question 9. All Democratic respondents will be presented with one of the variations of Question 10. Q9.1 (Condition A = Republican Control Group - No Demographic information provided): I am running to be the nominee for the [Democratic/Republican] Party in the general election this fall. I am running because I believe in America and want to do my part to help Americans thrive like they never have before. I pledge that I will work tirelessly to 185

203 achieve important goals like providing a quality education for our children, stimulating a strong economy, and a keeping our streets safe for our citizens. Q9.2 (Condition B = White Republican male candidate): The same message from Q9.1 will be presented along with a photograph of a fictional white male politician. Q9.3 (Condition C = White Republican female candidate): The same message from Q9.1 will be presented along with a photograph of a fictional white female politician. Q9.4 (Condition D = Black Republican male candidate): The same message from Q9.1 will be presented along with a photograph of a fictional black male politician. Q10.1 (Condition E = Democratic Control Group - No Demographic information provided): same message from Q9.1 will be presented with no accompanying photograph of a fictional politician. Q10.2 (Condition F = White Democratic male candidate): The same message from Q9.1 will be presented along with a photograph of a fictional white male politician Q10.3 (Condition G = White Democratic female candidate): The same message from Q9.1 will be presented along with a photograph of a fictional white female politician. Q10.4 (Condition H = Black Democratic male candidate): The same message from Q9.1 will be presented along with a photograph of a fictional black male politician. Note: Photographs of the three candidates to be used for both political parties: Part 4: Candidate Evaluation Note: all respondents will answer the same questions but party labels are applied to respondents of each party, respectively. For instance, Republicans will be asked to evaluate how good this candidate would be for the Republican Party and Democrats will be asked to evaluate how good this candidate would be for the Democratic Party. 186

204 Please provide us with some of your thoughts about this candidate. Be honest. There are no right or wrong answers. We are only interested in your opinion. Q11: Which of the following do you think best describes this candidate s political ideology? A11: Respondents will select one of the following buttons: Very liberal, Somewhat liberal, Closer to liberals, Neither liberal nor conservative, Closer to conservatives, Conservative, Very conservative Q12: I would support this candidate as the Republican Party s nominee for the fall election. A12: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q13: If elected, this candidate would support the interests of people like me. A13: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q14: How appealing do you think this candidate would be to independent voters and voters who belong to other political parties? A14: Respondents will select one of the following buttons: Very unlikely, Unlikely, Somewhat unlikely, Undecided, Somewhat likely, Likely, Very likely Q15: How confident are you that your impressions of this candidate are accurate? A15: Respondents will select one of the following buttons: Extremely confident, Very confident, Somewhat confident, Not very confident, Not at all confident Part 5: Personal Characteristics Block Q16: Which of the following best describes you? A16: Respondents select one of the following buttons: White/Caucasian, African American, Hispanic, Asian, Native American, Pacific Islander, Other Q17: Which is the highest level of education you have completed? A17: Respondents select education level from dropdown menu (Less than high school; High school/ged; Some college; 2-year college degree; 4-year college degree; Masters degree; Doctoral degree; Professional degree (JD, MD)) Q18: What is your gender? A18: Respondents select either Male or Female 187

205 Q19: What is your annual household income? A19: Respondents select income range from dropdown menu ($20,000-29,999; $30,000-39,999; $40,000-49,999; $50,000-59,999; $60,000-69,999; $70,000-79,999; $80,000-89,999; $90,000-99,999; $100, ,999; $110, ,999; $120; ,999; $130, ,999; $140, ,999; $150,000+) Part 6: Gender and Society Battery Next we would like to hear your opinions about modern American society. Please indicate the degree to which you agree or disagree with each of the following statements about men and women in America today. Q20: When women lose fairly to men in equal competition, they often claim discrimination. A20: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q21: Women do not need to be protected by men. A21: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q22: Women often exaggerate problems at work. A22: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q23: In a disaster, women do not necessarily need to be rescued first. A23: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Part 7: Political Preferences Q24: The Democratic [Republican] Party includes many types of people. In general, how would you describe yourself? A24: Respondents select either Strong Democrat (Republican) or Weak Democrat (Republican) Q25: Do you support, oppose, or neither support nor oppose the Tea Party? A25: Respondents select one of the following buttons: Support, Oppose, Neither support nor oppose 188

206 Q26: When it comes to politics, how would you describe yourself, and the following, as liberal, conservative, or neither liberal nor conservative? A26: For each of three entities ( You, Democrats, and Republicans ) respondents will select one of the following buttons: Very liberal, Somewhat liberal, Closer to liberal, Neither liberal nor conservative, Closer to conservatives, Somewhat conservative, Very conservative Q27: We would like to get your overall feelings towards political parties in America. On the scale below, ratings between 50 and 100 mean you feel favorable and warm toward the party. Ratings between 0 and 50 degrees mean that you do not feel favorable toward the party. A rating exactly at the 50 degree means you do not feel particularly warm or cold towards the Party. Where would you place yourself on this scale? A27: Respondents will use a sliding scale (0 to 100) to identify a level of favorability/warmth towards both the Democratic, Republican, and Tea Parties Q28: If you had to give President Obama a letter grade, where "F" means terrible and "A" means excellent, how would you rate his overall performance as president? A28: Respondents will use a sliding scale ( A+ to F ) to give President Obama a letter grade Q29: During the last few years, did you usually vote in national, state, and local elections, or did you usually not vote? A29: Respondents select usually voted or usually did not vote Part 8: Political knowledge battery Q30: What job or political office is held by Joe Biden? A30: Open-ended: Respondent inputs answer Q31: Whose responsibility is it to determine if a law is constitutional or not? A31: Respondents will select one of the following buttons: The President, Congress, The Supreme Court Q32: How much of a majority is required for the U.S. Senate and House to override a presidential veto? A32: Respondents will select one of the following buttons: Any majority, 51%, 60%, 67%, 75% Q33: Which political party would you say is generally more conservative than the other A33: Respondents will select one of the following buttons: Democratic Party, Republican Party 189

207 Part 9: Opinions about American Society Please indicate the degree to which you agree or disagree with each of the following statements about race in American Society. On average, African-American students get lower scores on standardized tests than do whites. How much of the difference in test scores: Q34: Occurs because most blacks do not have the chance to get a good education? A34: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q35: Can be explained by discrimination against blacks A35: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q36: Occurs because most blacks do not teach their children the values and skills which are required to be successful in school? A36: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q37: Occurs because most blacks just don't have the motivation or will power to perform well. A37: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q38: Is due to racial differences in intelligence? A38: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q39: Occurs because of fundamental genetic differences between the races? A39: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know On average, African-Americans have lower income and worse housing than white people. How much of the economic difference between blacks and whites: Q40: Occurs because most blacks do not have the chance to get a good education? A40: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q41: Can be explained by discrimination against blacks A41: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q42: Occurs because most blacks do not teach their children the values and skills which are required to be successful in school? A42: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know 190

208 Q43: Occurs because most blacks just don't have the motivation or will power to perform well? A43: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q44: Is due to racial differences in intelligence? A44: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Q45: Occurs because of fundamental genetic differences between the races? A45: Respondents will select one of the following buttons: A great deal, Some, A Little, None, Don t know Part 10: Candidate Recall Q46: Please think back to the candidate you evaluated earlier. What do you recall about the candidate? List anything that you remember. A46: Open-ended: respondent inputs answer Part 11: Debrief and Payment Code Thank you for participating! This study had you looking at a fictional political candidate within your own party to learn more about how people make decisions in elections. Important: In order to get credit for participating, you must input the following code into the Mechanical Turk payment code box. Note: a random code will be generated at the end of the survey which respondents must copy and paste into the Mechanical Turk web page to complete the survey. 191

209 Study 2: Electoral Effects of Counterstereotypical Cues Part 1: Eligibility Question Block Note: not all questions in the eligibility block directly concern eligibility. Questions 1, 2, and 6 are the pertinent queries. The others are included to minimize the likelihood that an ineligible respondent can go back and change an answer in order to make herself eligible after learning that she is not. These first few questions will be used to determine your eligibility for this survey. Please answer them honestly. Q1: What is your age in years? A1: Open-ended: Respondent inputs age Note: If respondent is under 18 years of age, the survey will end Q2: Are you a citizen of the United States? A2: Respondent selects yes or no Note: if respondent is not a citizen of the United States, the survey will end Q3: Were you born in the United States? A3: Respondent selects yes or no Q4: In what state do you currently reside? A4: Respondents select state of residence from dropdown menu Q5: Which of the following best describes your religion? A5: Respondents select religion from dropdown menu (Baptist any denomination; Protestant (e.g. Methodist, Lutheran, Presbyterian, Episcopal); Catholic; Mormon; Jewish; Muslim; Hindu; Buddhist; Pentecostal; Eastern Orthodox; Other-Christian; Atheist; Agnostic) Q6: Generally speaking, do you usually think of yourself as a Republican, a Democrat, an independent or what? A6: Respondent selects Republican, Democrat, Independent, Something else Note: If respondent is neither Republican nor Democrat the survey will end 192

210 Q7: Are you (Check all that apply): A7: Respondent select one or more buttons: Married, Divorced, Widowed, Separated, Single, Single parent Part 2: Personal Characteristics Block Thank you very much for taking this survey. Please tell us a little bit more about yourself. Q8: Which of the following best describes you? A8: Respondents select one of the following buttons: White/Caucasian, African American, Hispanic, Asian, Native American, Pacific Islander, Other Q9: Which is the highest level of education you have completed? A9: Respondents select education level from dropdown menu (Less than high school; High school/ged; Some college; 2-year college degree; 4-year college degree; Masters degree; Doctoral degree; Professional degree (JD, MD)) Q10: What is your annual household income? A10: Respondents select income range from dropdown menu ($20,000-29,999; $30,000-39,999; $40,000-49,999; $50,000-59,999; $60,000-69,999; $70,000-79,999; $80,000-89,999; $90,000-99,999; $100, ,999; $110, ,999; $120; ,999; $130, ,999; $140, ,999; $150,000+) Part 3: Political Preferences Q11: Which of the following options best describes how you view yourself as a [Democrat/Republican]? Note: Respondents will be sorted based on their answer to Question 6 so that they identify themselves within their own party A11: Respondents select either Strong Democrat (Republican) or Weak Democrat (Republican) Q12: Do you support, oppose, or neither support nor oppose the Tea Party? A12: Respondents select one of the following buttons: Support, Oppose, Neither support nor oppose Q13: When it comes to politics, how would you describe yourself, and the following, as liberal, conservative, or neither liberal nor conservative? A13: For each of three entities ( You, Democrats, and Republicans ) respondents will select one of the following buttons: Very liberal, Somewhat liberal, Closer to liberal, Neither liberal nor conservative, Closer to conservatives, Somewhat conservative, Very conservative 193

211 Q14: Generally speaking, how interested are you in what s going on in government and politics? A14: Respondents will select one of the following buttons: Extremely interested, Very interested, Moderately interested, Slightly interested, or Not interested at all Q15: During a typical week, about how many hours per week do you spending watching political media? A15: Open-ended: Respondent inputs number Q16: I would like to get your overall feelings towards political parties in America. On the scale below, ratings between 50 and 100 mean you feel favorable and warm toward the party. Ratings between 0 and 50 degrees mean that you do not feel favorable toward the party. A rating exactly at the 50 degree means you do not feel particularly warm or cold towards the Party. Where would you place yourself on this scale? A16: Respondents will use a sliding scale (0 to 100) to identify a level of favorability/warmth towards both the Democratic and Republican Parties Q17: During the last few years, did you usually vote in national, state, and local elections, or did you usually not vote? A17: Respondents select usually voted or usually did not vote Q18: How likely are you to vote in the upcoming 2014 Congressional elections? A18: Respondents select a button on a ten-point scale ranging from Extremely Unlikely to Extremely Likely Part 4: Candidate evaluation (including experimental conditions) Note: Respondents will be randomly assigned to learn about one of four fictional political candidates within their own party. Four candidates (A white male, a white female, a black male, and a candidate with no demographic information) will provide policy preferences which imply either ideological moderation or ideological extremism. All Republican respondents, then, will be presented with one of the variations of Question 19. All Democratic respondents will be presented with one of the variations of Question 20. Q19.1 (Republican Control A): Candidate with no demographic cues provides an ideologically moderate message: I am running to be the nominee for the Republican Party in the general election this fall. As a life-long public servant, I have always believed that our job in government is to provide basic services to the people so that they may be able to pursue the American Dream. I will work to make sure that our economy is strong and citizens have jobs, that our country is safe and protected, and that our children all have access to a quality education. The best way to achieve these goals is for everyone in government to come together and pursue reasonable, balanced public policy through compromise. The best public policy is made by taking the best ideas from both parties. 194

212 Q19.2 (Republican Control B): Candidate with no demographic cues provides an ideologically extreme message: I am running to be the nominee for the Republican Party in the general election this fall. As a life-long conservative, I have always believed that our job in government is to provide a few basic services to the public and otherwise stay out of the way so that everyone may be free to pursue the American Dream. I will work to make sure that the economy is free so that businesses can compete and grow and produce jobs. I will work to protect American citizens by ensuring that our military is the strongest in the world so it can protect our citizens. I will see that our children get the best education possible by supporting school voucher programs and reducing the negative influence of teachers unions so that school leadership is returned to the local level where it belongs. The best public policy is made by committing to core conservative ideals. As your candidate, I promise to do just that. Q19.3 (Moderate White Republican male candidate): The same message from Q19.1 will be presented along with a photograph of a fictional white male politician. Q19.4 (Extreme White Republican male candidate): The same message from Q19.2 will be presented along with a photograph of a fictional white male politician. Q19.5 (Moderate White Republican female candidate): The same message from Q19.1 will be presented along with a photograph of a fictional white female politician. Q19.6 (Extreme White Republican female candidate): The same message from Q19.2 will be presented along with a photograph of a fictional white female politician. Q19.7 (Moderate Black Republican male candidate): The same message from Q19.1 will be presented along with a photograph of a fictional black male politician. Q19.8 (Extreme Black Republican male candidate): The same message from Q19.2 will be presented along with a photograph of a fictional black male politician. Q20.1 (Democrat Control A): Candidate with no demographic cues provides an ideologically moderate message: I am running to be the nominee for the Democratic Party in the general election this fall. As a life-long public servant, I have always believed that our job in government is to provide basic services to the people so that they may be able to pursue the American Dream. I will work to make sure that our economy is strong and citizens have jobs, that our country is safe and protected, and that our children all have access to a quality education. The best way to achieve these goals is for everyone in government to come together and pursue reasonable, balanced public policy through compromise. The best public policy is made by taking the best ideas from both parties. Q20.1 (Democrat Control B): Candidate with no demographic cues provides an ideologically extreme message: I am running to be the nominee for the Democratic Party in the general election this fall. As a life-long progressive, I have always believed that our job in government is to provide important services to the people to help ensure that everyone 195

213 has what they need to pursue the American Dream. I will work to make sure that the government keeps an eye on the economy and makes sure that people from all walks of life can find good jobs. I will work to protect American citizens by ensuring that we engage and work diplomatically with countries around the world to prevent conflict. I will see that our children get the best education possible by supporting teachers and additional education spending in our classrooms so everyone has the resources they need to succeed. The best public policy is made by committing to core progressive ideals. As your candidate, I promise to do just that." Q20.3 (Moderate White Democratic male candidate): The same message from Q20.1 will be presented along with a photograph of a fictional white male politician. Q20.4 (Extreme White Democratic male candidate): The same message from Q20.2 will be presented along with a photograph of a fictional white male politician. Q20.5 (Moderate White Democratic female candidate): The same message from Q20.1 will be presented along with a photograph of a fictional white female politician. Q20.6 (Extreme White Democratic female candidate): The same message from Q20.2 will be presented along with a photograph of a fictional white female politician. Q20.7 (Moderate Black Democratic male candidate): The same message from Q20.1 will be presented along with a photograph of a fictional black male politician. Q20.8 (Extreme Black Democratic male candidate): The same message from Q20.2 will be presented along with a photograph of a fictional black male politician. Note: Photographs of the three candidates to be used for both political parties: 196

214 Part 4: Candidate Evaluation Note: all respondents will answer the same questions but party labels are applied to respondents of each party, respectively. For instance, Republicans will be asked to evaluate how good this candidate would be for the Republican Party and Democrats will be asked to evaluate how good this candidate would be for the Democratic Party. Please provide us with some of your thoughts about this candidate. Be honest. There are no right or wrong answers. We are only interested in your opinion. Q21: Which of the following do you think best describes this candidate s political ideology? A21: Respondents will select one of the following buttons: Very liberal, Somewhat liberal, Closer to liberals, Neither liberal nor conservative, Closer to conservatives, Conservative, Very conservative Q22: I would support this candidate as the Republican Party s nominee for the fall election. A22: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q23: If elected, this candidate would support the interests of people like me. A23: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q24: How appealing do you think this candidate would be to independent voters and voters who belong to other political parties? A24: Respondents will select one of the following buttons: Very unlikely, Unlikely, Somewhat unlikely, Undecided, Somewhat likely, Likely, Very likely Q25: How confident are you that your impressions of this candidate are accurate? A25: Respondents will select one of the following buttons: Extremely confident, Very confident, Somewhat confident, Not very confident, Not at all confident Part 5: Political knowledge battery Q26: What job or political office is held by Joe Biden? A26: Open-ended: Respondent inputs answer Q27: Whose responsibility is it to determine if a law is constitutional or not? A27: Respondents will select one of the following buttons: The President, Congress, The Supreme Court 197

215 Q28: How much of a majority is required for the U.S. Senate and House to override a presidential veto? A28: Respondents will select one of the following buttons: Any majority, 51%, 60%, 67%, 75% Q29: Which political party would you say is generally more conservative than the other? A29: Respondents will select one of the following buttons: Democratic Party, Republican Party Part 6: Opinions about American Society Please indicate the degree to which you agree or disagree with each of the following statements about yourself and modern American Society Q30: Most Americans are unselfish towards others. A30: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q31: Most Americans tend to be disorganized. A31: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q32: Americans tend to be more interested in sports than in art and culture. A32: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q33: When women lose to men in fair competition, they typically complain about being discriminated against. A33: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q34: The world would be better if women support men more and criticize them less. A34: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q35: The more women advance in the business world the better. A35: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q36: Most Americans tend to have a forgiving nature. A36: Respondents will select one of the following buttons: Strongly disagree, 198

216 Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q37: By and large, Americans tend to be more assertive than other people around the world. A37: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q38: I prefer new and predictable experiences over fun but routine ones. A38: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q39: I laugh more when I watch a comedy with others than when I watch it alone. A39: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q40: I am not always the person I appear to be to others. A40: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q41: I would not change or modify my opinions in order to please someone else or win favor. A41: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q42: Even if I am not enjoying myself, I often pretend to be having a good time. A42: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q43: Americans tend to worry and stress more than they need to in life. A43: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q44: People today are more interested in sharing opinions about events than learning about all relevant facts about them. A44: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q45: I would prefer to spend an evening in with a good movie or book over going out to a bar or club. A45: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree 199

217 Q46: Discrimination against racial minorities is no longer a problem in the United States. A46: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Q47: Affirmative action programs on colleges campuses constitute reverse discrimination. A47: Respondents will select one of the following buttons: Strongly disagree, Disagree, Somewhat disagree, Neither agree nor disagree, Somewhat agree, Agree, Strongly agree Part 7: Candidate Recall Q48: Please think back to the candidate you evaluated earlier. What do you recall about the candidate? List anything that you remember. A48: Open-ended: respondent inputs answer Part 8: Debrief and Payment Code Thank you for participating! This study had you looking at a fictional political candidate within your own party to learn more about how people make decisions in primary elections. Important: In order to get credit for participating, you must input the following code into the Mechanical Turk payment code box. Note: a random code will be generated at the end of the survey which respondents must copy and paste into the Mechanical Turk web page to complete the survey. 200

218 APPENDIX B IRB DOCUMENTS STUDY 1 The follow pages include all documents submitted to the Institutional Review Board (IRB) at the University of Pittsburgh for approval of Study 1. The project was granted IRB approval on March 26,

219 202

220 203

221 204

222 205

223 206

224 207

225 208

226 STUDY 2 The follow pages include all documents submitted to the Institutional Review Board (IRB) at the University of Pittsburgh for approval of Study 2. The project was granted IRB approval on November 27,

227 210

228 211

229 212

230 213

231 214

232 215

233 216